Connecting Autonomous Systems with NextDroid, Part 2

In Episode 17 of The Connext Podcast:

Related Content:

• [Case + Code] Accelerate Autonomous Car Development
• [Press Release] AUTOSAR Releases Latest Adaptive Platform Adopting the DDS Standard
• [Press Release] NextDroid Selects RTI’s Connectivity Technology for Its State-of-the-Art Autonomous Systems
• [Podcast] Connecting Autonomous Systems with NextDroid, Part 1
• [Webinar] Secrets of Autonomous Car Design
• [Whitepaper] The Secret Sauce of Autonomous Cars
• [Whitepaper] ISO 26262 Compliance Using Approved Software Components for Road Vehicles

Podcast Transcription:

Steven Onzo: Hi, and welcome to Episode 17 of the Connext podcast. I'm Steven Onzo, producer of the Connext podcast, and today we'll be continuing an interview recorded in December 2017 with Lacey Trebaol and Daniel Gandhi, Director of Autonomous Vehicles at NextDroid. In part two of this two part series, we'll talk about the early stages of Dan's career, where he earned a triple major in physics, computer science, and math, and how he's applying that background to solve the complex obstacles of automotive autonomy.

Steven Onzo: We'll also learn the difference between passive and active safety and how smart systems can keep passengers secure. We hope you enjoy this episode.

Lacey Trebaol: Alright, so I tried looking at your LinkedIn profile. There is not much there. It was like, it was anemic. I was going, "Is this the right person?", at first. But then I saw it was Stanford, I was like "This has to be the right person!"

Daniel Gandhi: Right.

[0:53] Dan’s academic background and tech interests prior to NextDroid

Lacey Trebaol: So if you could talk just a little about your academic background maybe? Start with what you studied in for your undergrad and some of the projects you thought were cool and informative, and technologies you were really into, and then lead up to, "And then you ended up here."

Daniel Gandhi: Sure. So, basically, I would say that what I'm generally interested in are complex, real world problems. Problems that bridge different disciplines and I feel like those are areas where we, as society, have a lot of room to grow. We've tended to pigeonhole ourselves into specific disciplines and basically mastered them. But as you get to problems where you are having to draw on a lot of different disciplines and try to combine them in ways to solve problems that are really tactile.

Lacey Trebaol: Which is like real creativity.

Daniel Gandhi: Exactly. There's a lot of room to have impact, there's a lot of room to grow there. And so, I think that's a common thread that you'll see throughout my background, is an affinity towards that.

Daniel Gandhi: So, during my undergrad, I went to Brandeis University and I triple majored in physics, computer science and math.

Lacey Trebaol: That's awesome.

Daniel Gandhi: And if you think about how those disciplines fit in together, they fit into that framework. You're talking about understanding how things work, physically, through the physics element of it. And then software's basically what makes the world go around these days, so ...

Lacey Trebaol: Math is how to describe it.

Daniel Gandhi: Exactly.

Lacey Trebaol: Yeah.

Daniel Gandhi: Math, you can model anything that you need to.

Lacey Trebaol: Without it, you can't do this, because you can't describe it. Computers operate on that.

Daniel Gandhi: Right.

Lacey Trebaol: You need to speak that language, right? It's-

Daniel Gandhi: Exactly. And you start seeing connectivity through math, when you start seeing how things gel together in ways that you didn't expect. Because math is the underpinning of so many different disciplines, you can start seeing how they all relate to each other and how you can extract basic tenets of how you should operate any discipline, based on the mathematical principles that feed into it.

Lacey Trebaol: I also like that with math, it's a great evaluator of truth, because it's well-formed or it's not. And so, you're getting this ... Or, especially early on, before you commit a bunch of time to something, it's like, is this a solid thing? Does this concept make sense? And you can apply that to so many things, just like that methodology, I guess, the rigor behind it and stuff. So yeah, I totally get those three majors. That's an awesome set of majors.

Daniel Gandhi: It was not intended. It was accidental.

Lacey Trebaol: You accidentally stumbled on to three majors?

Daniel Gandhi: Yes. My plan was always to major in physics and I was going to double minor, which is not super uncommon. But, I essentially front-loaded my schedule a lot, and that meant that I was taking a lot of classes that would cover majors in math and computer science. And I was in senior year and advisors were like, "If you took one class in this topic, you could get a major instead of a minor." And same thing with the other-

Lacey Trebaol: That's so funny.

Daniel Gandhi: And so I basically added a couple of more classes to my schedule and-

Lacey Trebaol: Ended up with an accidental triple major.

Daniel Gandhi: Right.

Lacey Trebaol: That's awesome.

Daniel Gandhi: In my undergrad, I did do a senior thesis project, and I was looking for something interesting to work on, that, again, would kind of fall in line with the three things I was talking about. And we have a lab at Brandeis that was called the Graybiel Lab. They partner with NASA and they look at how humans could operate for a long term in space. And so one of the experiments that they'd do is, they have a rotating room. They put you in a rotating environment and that is how you simulate artificial gravity for a long space mission.

Lacey Trebaol: Right.

Daniel Gandhi: But, because you're creating art gravity through rotation, you create side effects.

Lacey Trebaol: Through mechanical means, right.

Daniel Gandhi: So, you essentially have Coriolis forces that are happening all the time. And so they ran a series of interesting experiments. They sit you down in the room and they have you move your arm to try to touch a target. But the room is darkened, so you can't really see anything except the target. And as soon as you start moving your hand, the target disappears. So, you're essentially saying, "Oh, I knew where this was."

Daniel Gandhi: Just like you'd reach for something without looking at it. Like, you see something, you start reaching, you look away, they're doing that. And your body has mechanical feedback mechanisms that tell you that you're on track.

Lacey Trebaol: Right.

Daniel Gandhi: And so, they would have you move your hand in this manner, and they have a motion-tracking system in the room that's IR, that's tracking higher hand action moves. And what happens is, in a rotating environment, you miss. Because the room is-

Lacey Trebaol: Because your kinesthetic awareness is not going to compensate for-

Daniel Gandhi: Well, it actually will, but it takes time.

Lacey Trebaol: Oh, that's right, you have to learn it.

Daniel Gandhi: So, as you move, you'll miss, and your body will be like, "Wait, we didn't end up where we were supposed to be." And so, if you do it again, you'll actually adapt. Until it converged back to the correct line.

Daniel Gandhi: If I put an object in your hand, what happens is, now you have a new mass and you haven't understand how that mass will basically react to the environment. And so you'll miss again. And as you keep moving it, you'll adapt. Where it gets really interesting is, if I start the rotation up and I have not let you adapt to anything, and I put an object in your hand and you move and you miss, you will slowly adapt to habit and move correctly. And if I take the object out of your hand, your hand has already adapted.

Daniel Gandhi: So, basically your body is able to go and decouple what correction was required for itself, and what correction was required for whatever you were holding, in one fell swoop.

Lacey Trebaol: Wow.

Daniel Gandhi: And so, they were trying to understand how that could be possible. What types of feedback your hand would experience from objects. And so my thesis project was to basically try to build a device that you could hold, that would track pressure across your hand as you moved an object around. And that way they would be able to couple the motion of the hand, through the motion tracking system, as well as the forces that the hand experienced at different points, so that they could try to couple ... You know, there are certain nerves at certain places in your hand, and all the different pads that exist, and how do they then couple with your central nervous system to create the correction.

Lacey Trebaol: Those are fun things to think about.

Daniel Gandhi: Yes.

Lacey Trebaol: So they were just holding a ball with sensors in it at that point.

Daniel Gandhi: Yeah. It was actually a cylinder and it had sensors on the outside of it, so when you grasped it, it could basically report where you were feeling certain forces as you moved it.

Lacey Trebaol: Awesome. So that was the undergrad work?

Daniel Gandhi: Yes.

Lacey Trebaol: And then, you were like, "I've not had enough of school. I have my triple major and I need more." ?

Lacey Trebaol: So did you take a break between undergrad?

Daniel Gandhi: I didn't.

Lacey Trebaol: You went straight.

Daniel Gandhi: I went straight to do a Master's in electrical engineering at Stanford. I was looking at it more from ... Physics is the theoretical framework, then engineering is going to be the application of it. And picking electrical engineering, it bridges. You can be very computer science-y in electrical engineering. You can be a lot more towards hardware. And so, I picked something in the middle that I could span space. And that's also where I started to focus more on robotics and more on control as specific applications.

Daniel Gandhi: Basically seeing that robotics and automation is the epitome of multidisciplinary. You have lots of different things that come together to make a robot, but then the robot interacts in the real world, and now you have everything that it could possibly interact with. And it's a field that there's a lot of momentum in right now. So that means that there's a lot of opportunity, while there are things that you can work on that also meet that criteria, if there isn't a lot of investment, it's still hard to make an impact. And so, robotics is something that looked like it was going to have an inflection point.

Lacey Trebaol: It was appropriately poised for you.

Daniel Gandhi: Yes.

Lacey Trebaol: Yeah. It works really great in the lab, right?

Daniel Gandhi: Right.

Lacey Trebaol: And then you put it in actual context, the thing has to operate on an... emergent behaviors.

Daniel Gandhi: Right.

Lacey Trebaol: Things happen.

Daniel Gandhi: And you have, you can see that in the progression. I mean, robots have been primarily deployed thus far in factory settings. Where you have-

Lacey Trebaol: Industrial automation type.

Daniel Gandhi: Exactly. You have a lot of control over the environment. You have cages that make sure that everything that the robot works on is restricted to what they are.

Lacey Trebaol: You really do.

Daniel Gandhi: You don't have ... As soon as somebody walks into a work cell, the robots stop.

Lacey Trebaol: Do they?

Daniel Gandhi: Yeah.

Lacey Trebaol: Always?

Daniel Gandhi: Pretty much.

Lacey Trebaol: Really? Like it's so huge and heavy, so, it's a huge safety-

Daniel Gandhi: Exactly.

Lacey Trebaol: OSHA would probably not be too thrilled with ...

Daniel Gandhi: Right. So you're seeing, from the research side, you're seeing a lot of people trying to work on collaborative robots. But now you have this problem of how do I have such a large heavy thing, with this wild card in its environment, which is a person. And so, while you're trying to ensure that a person is safe ... How do you have a robot and a person work together?

Lacey Trebaol: Right. Not only safe, but they have to achieve some objective.

Daniel Gandhi: Right.

Lacey Trebaol: Right.

Daniel Gandhi: It has to be productive collaboration.

Lacey Trebaol: We cannot just exist in the same space and call it a day. You're trying to actually do something.

Daniel Gandhi: Right.

Lacey Trebaol: So what did you do, when you were at Stanford, then?

[10:12] An introduction to Robotics at Stanford

Daniel Gandhi: Mostly I did coursework there in sort of traditional electrical engineering pieces: controls, robotics. One of my classes was Experimental Robotics, where we looked at trying to come up with an interesting task for a robotic arm. And the professor was always ... I was looking for students to come up with some novel project that would kind of push limits. And you worry you really wouldn't know if it would work well. Just to see if you could come up with something cool to do.

Daniel Gandhi: And we had this old industrial arm that was a six degree of freedom arm. We essentially took on the project of trying to get it to throw a Frisbee. Which is ...

Lacey Trebaol: The flick ... Oh, wow. Yeah.

Daniel Gandhi: And it was pretty challenging. By the end, the project had to shift to something that the robot could actually attain. So, industrial robotic arms are not known for being particularly agile.

Lacey Trebaol: No, or graceful.

Daniel Gandhi: Right. So the amount of ... Just all the complex motion you have to do to get a Frisbee to move is pretty tough. So we had tried experiments with it, and you have to get the Frisbee to rotate as well as get thrown. So we tried a lot of different ways of doing and we actually ended up with something where ... There are these toys that you basically ... They have kind of a small helicopter blade. There's a motor underneath, and it kind of spins them up, and then it launches them up, and they kind of pop up and float, through the air.

Lacey Trebaol: Yeah.

Daniel Gandhi: So, inspired by that, we essentially took a fabric Frisbee, which would be light enough that we could-

Lacey Trebaol: Like the nylon ones, kind of?

Daniel Gandhi: Yeah. Right. And we attached the mechanism to it that would make it mimic this toy. So then you could put a motor on, you could spin it up at speed, and that would create a lot of stability in the Frisbee itself. And using the same mechanism, basically what it would do is when you'd start the motor, it would pop the Frisbee off. And that means that the arm could actually start doing a throw with this Frisbee. When you want to release, it'd create this pop, and there'd be immediate separation between the robot and the Frisbee.

Lacey Trebaol: Right. So it would lift and it would be ... Yeah.

Daniel Gandhi: Right. And so then you could actually throw a Frisbee across a room to somebody and we actually ... For kicks, we tied in a camera to it. So, it would just track the color red, the Stanford color, it's cardinal. So, it would actually find somebody wearing red and it would throw the Frisbee towards them.

Lacey Trebaol: That's funny.

Daniel Gandhi: So ...

Lacey Trebaol: You created the targeting system?

Daniel Gandhi: With a Frisbee, sure.

Lacey Trebaol: And then when did you graduate?

Daniel Gandhi: I graduated in 2008. I'd also say that, while I was in school, I interned at a lot of places-

Lacey Trebaol: Really?

[12:52] Technical projects Daniel worked on early in his career

Daniel Gandhi: I started interning, maybe when I was like, 15 or something. And they were also across different industries. So, the first place was in the RF microwave realm. I was at a company that made little RFICs that end up in your cell phone and that type of thing. This was in the early 2000s, so they weren't as pervasive. But, these are the pieces that kind of enable what we have as wireless communication being so pervasive today.

Lacey Trebaol: Yeah.

Daniel Gandhi: And so they were working on ... They had come out of the defense side, and then they saw that commercial applications were starting to become more attainable, and they were moving into that space. So when I was there, I worked on ways to basically create automated tests for their chips, where tests can be kind of problematic. You can very easily have human error in trying to test it, that gives you wrong results.

Daniel Gandhi: And so, they were looking to create test stands that would control all the instruments and inject the right signals. You could have chips that come out and-

Lacey Trebaol: We know they work. Or, we know what the issue is.

Daniel Gandhi: Right. You know their performance and you know that they work properly and that type of thing.

Lacey Trebaol: Yeah.

Daniel Gandhi: And then, I continued there and I actually made, basically, engineering tools that they would post on their website. They would kind of draw customers in to help them calculate different aspects of what their jobs would be.

Daniel Gandhi: So, the first one that I really did was, there's a device called a mixer. It's pretty much used everywhere. In a communications system, you use it to take something that's a low frequency signal and step it up, so that you can broadcast it, and then step it back down, so that you can process it. And they have all these side effects. You create all this spurious signals along the way. So this was a tool that actually mapped it out for you. So you could go and it would create a chart that would tell you, here's all the different spurious symbols that you're going to end up with, and what power they're going to be at. And that way, you could ... See, as I say, I'm going to operate this frequency range from my transmit and this frequency range for my base frequency and this frequency range that I'm going to be sweeping over to step it up and down ... Where am I going to run into the problem? So, it would help you map that out.

Lacey Trebaol: Cool. So you wouldn't have to measure it yourself. It would actually calculate it?

Daniel Gandhi: Correct. And you could plan it, basically, so as part of your system design you would use this as a step to make sure that you didn't have some oversight that basically made your system not operational by the time you built it.

Lacey Trebaol: Like, what do you mean I don't have enough power to get it to do this, this way? What do you mean these frequency ranges aren't working right for this?

Daniel Gandhi: Or you have like, essentially, a spurious signal that sits right on top like a real one. So close that you can't filter it out in any way, and now the whole system just doesn't behave the way that it needs to.

Lacey Trebaol: Cool. So you did that ...

Daniel Gandhi: Yep. And then I went to a company that was making circuit simulation software. So, they essentially were looking at trying to make a circuit simulator that could scale to huge circuits. So if I want to model RAM, where you have hundreds of thousands of transistors that all have to operate together, right now what would happen is that people would model small segments of it.

Lacey Trebaol: Right, and then they would try to extrapolate.

Daniel Gandhi: Exactly.

Lacey Trebaol: From that.

Daniel Gandhi: How that would scale. Here, that size now, is always a magnitude large, because you could more efficiently simulate larger circuits and at higher frequencies. And so, this was a small team of PhDs out of MIT that were trying to create a disruption in the market where they could topple major industry competitors. And so, I worked with them on modeling circuit components that their customers would need to build up their systems.

Lacey Trebaol: Cool.

Daniel Gandhi: And then, after that, I went to the MITRE Corporation. Essentially, MITRE is a non-profit that is set up to provide system engineering resources to the government.

Lacey Trebaol: Where.

Daniel Gandhi: The government can't bring in staff of that ilk, so Miter is the in between, between a primary defense contractor and the government, there's somebody who can check and create specs and what not.

Daniel Gandhi: What they also do is, they have a lot of their own internal research projects, where they're trying to stay up on technology. And I was there at a period when they were very interested in netted sensors. So, can you go and create networks of relatively cheap sensors and have them communicate so that you can have capability over large distances at lower costs?

Daniel Gandhi: An example of a project was, we were trying to create a sensor fence, where you could put an aerial fence over the US border. And you want to see ... I can detect any sort of low-flying aircraft that are trying to cross the border. Normally, you'd have expensive radar system that you would deploy. But, because of the curvature of the earth and terrain masking, you can't put one of those frequently enough to cover the entire border. But they have extremely sophisticated capabilities.

Daniel Gandhi: So now, can I try to replace those capabilities with something simpler but that I can just kind of spread everywhere?

Lacey Trebaol: Yeah. But netted out, so I can correlate and integrate and take all my data and-

Daniel Gandhi: Exactly.

Lacey Trebaol: Paint a useful picture.

Daniel Gandhi: What we did is we actually took a radar that you'd have on a boat. Like your small craft, like a 4,000 dollar type of radar system that you'd put on there. And we modified it, so that instead of spinning, it would create a beam, almost like a radar trip line in the sky. And then we also put acoustic microphone array and so now you would essentially ... If anything cut the line, you could detect that that happened and you'd have microphone array that would tell you the direction that it generally was-

Lacey Trebaol: So you could do DF, because you had enough, yeah.

Daniel Gandhi: So you could basically figure out that there's some object there, and you could get an acoustic signature on it. We basically built these sort of homegrown nodes and we would go to airports and we would set up near the runways and we'd put these nodes down, and it would find aircraft as they crossed the line. The idea was that, these are relatively cheap, you could spread out. And then you'd have more expensive, centralized stations that would have things like high resolution cameras and thermal and visual. And then they would then be able to train themselves to latch on to whatever this unknown object was. And that could then be pre-processed, and you've an operator somewhere who might be, you know, surveilling hundreds of miles of-

Lacey Trebaol: On a ground station or in an ISR platform that's in the sky, or ...

Daniel Gandhi: And so now you're connecting all the data together and you're trying to replace the capability of something that was much more expensive and covered far less area.

Lacey Trebaol: All right, so that was internship number four. You had some crazy internships.

Daniel Gandhi: Yeah.

Lacey Trebaol: So you ended up then, after that one, where?

Daniel Gandhi: I started at MITRE when I was at the end of my undergrad, and I kept going back to MITRE while I was at Stanford, working projects, during breaks or things like that. And then after Stanford, then I got my first time full-time job, so-

Lacey Trebaol: Which was where?

Daniel Gandhi: That was at General Motors.

Lacey Trebaol: Awesome. So what did you get to do there?

[20:06] Achieving active safety in autonomous driving systems

Daniel Gandhi: I worked there on active safety and autonomous driving systems. Before the modern hype and push in that area, this was the path that the OEMs were already on.

Lacey Trebaol: So what's active safety?

Daniel Gandhi: So, active safety ... Passive safety is what people think of as safety, currently. Basically, passive safety is everything that protects you after a crash has occurred.

Lacey Trebaol: Airbag.

Daniel Gandhi: So, your airbag, your seat belts, your explosive pretensioners, your ... just the structural elements in trying to absorb energy. All of that is passive safety. And if you look at accident statistics, injury rates, we're kind of plateauing in how much you can do with materials in some form. Once an accident has occurred, how much can you do to save somebody's life or reduce the amount of injury?

Daniel Gandhi: Europe, for example, has been having a lot of problems with density of pedestrians. And you have car accidents with pedestrians ... Sure, I can create a cage around the driver, but what do I do about the guy outside the car? And so people have experimented with like, airbags on the outside-

Lacey Trebaol: Oh, I've seen those. I've seen the cars with those. I thought was a joke. It ended up in my Reddit feed once, and it was like, "Is this a meme? Are they being funny?" Like no, it was an experiment.

Daniel Gandhi: And you also notice that cars have a higher belt line, a higher hood. A lot of that is around the fact that when you have really low cars, they kind of cut your legs out from under you if you get hit by one. And you end up in the windshield. And if you get hit by the windshield, it could kill you. But if you can get a person over the car-

Lacey Trebaol: Then they balance it up-

Daniel Gandhi: They balance it, it's a bit more likely to survive.

Lacey Trebaol: It displaces energy out, yeah.

Daniel Gandhi: You don't want to have all the energy just come together-

Lacey Trebaol: In one impact.

Daniel Gandhi: With the person trapped in the windshield on the impact.

Daniel Gandhi: So, you're seeing these things that people don't necessarily recognize as safety elements, but they're around safety regulations for trying to protect people.

Lacey Trebaol: You're like, "He bounced three times. That's good."

Daniel Gandhi: Right, exactly.

Lacey Trebaol: Physics, guys!

Lacey Trebaol: Is there regulation that says they have to do certain things like this? Like, heights for hoods and ...

Daniel Gandhi: I'm not sure what the regulations actually are. I wouldn't be surprised if they're doing crash dummy tests in locations with the dummy outside the car, to see what happens. Another piece with the passive safety side to it, is, if you think about how much a vehicle weighs today, versus how much it did. A car from 15 years probably weighed a thousand pounds less than it does today. If you look at what the power of an engine is today, the efficiency of an engine is, they're countering each other. I'm adding more structure to the car, I'm making the car heavier so that it can be safer to meet increasing safety regulations.And then I have to have an engine that's more efficient and more powerful to pull that car, and that's why-

Lacey Trebaol: That's why it's heavier.

Daniel Gandhi: And that's why you actually see fuel economy not explode, despite the fact that technology is becoming way more capable. The fact that you can have a Corvette cruise on a highway and get over 30 miles per gallon, there's like a huge amount of technology that went to make that happen. And then you can hit the gas and have 450 horsepower ... It's a huge feat. And when you say, "Oh well how come I don't have common hundred mile an hour per gallon cars?" It's because we've also increased the weight of cars. Active safety starts to address all those things. We start to say, "Can I prevent the accident from occurring? Can I mitigate the accident? Can I remove energy prior to the accident?"

Daniel Gandhi: And so, these are the technologies that you're seeing in the car. Simple technologies could just be warnings to drivers. I'm going to have a -

Lacey Trebaol: The beeps when you approach something too close.

Daniel Gandhi: Exactly. So you have a forward collision alert, you'll have a lane departure warning. And you're seeing that these things are becoming more standard in cars.

Lacey Trebaol: Yeah, my car has it, standard. The wheel will vibrate if I am going to travel out of my lane. I can take my hands off it, but ... And it's a Honda CRV. They were like, when I test drive it, "You can take your hands off." I was like, "I'm good!" That was the most unnatural thing in the world to ever do, was to get on a freeway, be driving about 60 miles an hour, and take your hands of a wheel and trust it.

Daniel Gandhi: Yes. Right.

Lacey Trebaol: Yeah. That was a trip. But yeah, so they're becoming more standard because it's definitely not an option I was seeking. But, I love that I have some things that I can auto set. Like one of mine is, at certain speeds, it will detect if you keep going like this, judging by what is in front of you, and it senses you're going to hit it, so it gets angry at you. That's what my daughter says, she goes, "It yells at you." And it'll apply the brakes. It will start pumping brakes for you.

Lacey Trebaol: So you don't do the mistakes like ... I was talking to my friends and she was saying with it that she was in a Starbucks drive through, and she was on her phone and let her foot off the brakes, and the car stopped her from hitting the car in front of her. So, these little accidents, right, I mean, they account in those statistics that you were talking about. Whiplash can happen. All these bad consequences, the cars are not ... they're overriding user error, I guess.

Daniel Gandhi: Right.

Lacey Trebaol: Which is nice.

Daniel Gandhi: And then you look at the opposite side, and you look at what kills people in car accidents. And, you're talking about high speeds, you're talking about lack of attention for a short period of time. How do you start covering those situations?

Daniel Gandhi: So, you could basically attack the problem from both sides and say, I'm going to make it so that you don't run into these minor problems that is really easy for the car to cover safely. And then you can look at these more complex scenarios and say-

Lacey Trebaol: The pressure with the time scale is being so much shorter.

Daniel Gandhi: Right.

Lacey Trebaol: Yeah.

Daniel Gandhi: And the car doesn't have the capability to necessarily do the optimal thing. It just doesn't have the information and processing to do that. But what it can do is say, "You know, we're going to crash into something." I can see that we're going to crash into something at this point. If it was going to be a lethal accident, we can take that down and we'll turn it into a serious injury, and a serious injury into a minor injury, and so on and so forth.

Lacey Trebaol: Right.

Daniel Gandhi: And you can do that by basically just trying to shave off speed of impact. Like, right at the very end, you know that a hit is going to happen. And the reason that you wait until that point is because the driver can actually avoid the accident entirely far later than the car can. Because you can steer. So, you could steer away to avoid something, but steering away is a very dangerous act for a car to take on its own. It doesn't know where it's steering into, so-

Lacey Trebaol: Yeah, it doesn't know that there's people on the side of the road, and, yeah ...

Daniel Gandhi: Right, is somebody running up from the back? Is there a guardrail? Is there a tree that it may not be able to detect super easily?

Lacey Trebaol: Are you crossing a bridge? Yeah.

Daniel Gandhi: So, it takes a conservative action, but one that can save lives in and of itself.

Daniel Gandhi: But, all these small things as well. So, for example, when you're ... I don't know how about the Honda system, but when you're looking at what options are available, as you know that accidents are looming, you can actually start preparing the car. So, for one thing, some vehicles have seat belts that have motors in them, so they can actually take up the slack and pull you back into the seat so you're anchored to the car in advance. They are-

Lacey Trebaol: So that you don't whip out.

Daniel Gandhi: Right.

Daniel Gandhi: During the crash, you ideally want to be as anchored as possible.

Lacey Trebaol: Right.

Daniel Gandhi: And, right now one of the things that they do to do that is there's an explosive pretensioner in the seatbelt, where it'll fire and it'll yank the belt super hard to pull you back into the seat before the airbag goes off. That is ... Anytime anything explodes in the car, it's a huge cost to repair, so they can't be wrong, which was -

Lacey Trebaol: And there isn't no injury associated with it.

Daniel Gandhi: Right, yeah. Especially with senior citizens, you have issues of breaking the sternum, that type of thing, when you have the types of force you're talking about during a crash.

Lacey Trebaol: Internal bleeding.

Daniel Gandhi: So, given those, this allows you to be a little more active. I can start pulling out slack, I can anchor you into the car, and I could be wrong, because it's a motor, it can let go. I don't have to have a reaction where I have to be right, and in order for me to be guaranteed to be right, I have to make it really late. And because it's really late, I have to be very aggressive.

Lacey Trebaol: Right.

Daniel Gandhi: I can actually soften that whole process. So, something like seat belts to get somebody in place, you can actually prepare the braking system so that you have more braking authority when you hit the brake. There are all these different things that you can do that-

Lacey Trebaol: That the car acts as a system and balances these actions.

Daniel Gandhi: Right.

Lacey Trebaol: So that there's the goal of minimizing the damage, right? As other things are applied, or other things are discovered in the system, it can weigh them appropriately.

Daniel Gandhi: Yes.

Lacey Trebaol: And then compensate for that.

Daniel Gandhi: So, it can go and say, I'm not sure that there's something wrong. I'm warning the driver. I want to make sure that if the driver thinks that there's something wrong that they have as much authority as possible to do something about it, so I'm going to prepare everything that I can-

Lacey Trebaol: It's like a heightened sense of alert.

Daniel Gandhi: Exactly.

Lacey Trebaol: Yeah.

Daniel Gandhi: And if they don't do anything, then I'll take my conservative action at the last minute, to try and do something.

Lacey Trebaol: And then no harm, no foul, kind of. Because you're not doing the crazy action that's not reversible. "We didn't blow an airbag in your face, we just pumped your brakes a few times.", yeah.

Daniel Gandhi: Right. The automatic braking systems the aggressive ones. The ones that are ... Basically in the industry, it's a little hard to tell from a consumer standpoint, but there are systems that just have some minimal intervention, and there are systems that intervene substantially.

Lacey Trebaol: Aggressively.

Daniel Gandhi: Exactly. And that's based on how confident the manufacturers are in terms of-

Lacey Trebaol: Really.

Daniel Gandhi: In terms of how ... One, their philosophy for how they want to move into this new space as well as ... For example, if I have a system that automatically brakes and I'm on the highway at 60 miles an hour, and it falsely automatically brakes, an aggressive system can apply, basically, maximum braking force. Which will put you out like a g of decel. Which is way more than a person actually ever-

Lacey Trebaol: Needs to be experiencing. Or would experience on their own through, yeah ...

Daniel Gandhi: They never execute it, a braking maneuver of that magnitude. So, imagine you're going down the highway and suddenly, out of nowhere, you experience a one g decel. There's traffic coming up behind you and they're not used to braking that hard. You're going to get hit from behind, and you're not only going to be able to intervene because your...

Daniel Gandhi: Yes. You're not going to state. Because you're just startled at this point, it's too, "What in the world is going on in my car?", to intervene.

Daniel Gandhi: So, because of that, you may, as a manufacturer, you may not want to be so aggressive. Or you may want to wait until you're very sure. In which case, the impact the system has is minimized. You might say, "Oh, we've developed a system that's very capable. Because it's so capable we can be sure earlier and we can intervene sooner." And so, you have, from a consumer standpoint, you basically see that these features exist in the car. You don't know how well they perform or what they actually do. And so, you have people like the Insurance Institute of Highway Safety and National Highway Transportation Safety Administration, starting to bake these things into crash tests.

Daniel Gandhi: As times goes on, you'll start seeing, just like you see, what's the front crash rating, you're going to see what-

Lacey Trebaol: These responsiveness ratings and stuff are.

Daniel Gandhi: Exactly.

Lacey Trebaol: And then understanding what the consequences of being on either end of a spectrum for a responsiveness thing would be ... Right.

Lacey Trebaol: So it's pros and cons from both.

Daniel Gandhi: Yeah.

Lacey Trebaol: That's so neat. I had never thought of that. Just scary, because I drive a car.

Lacey Trebaol: Okay, so how long were you at GM?

Daniel Gandhi: About three and a half years.

Lacey Trebaol: And then you went to ...

Daniel Gandhi: I came to a company that's in Cambridge, Massachusetts, Vecna Technologies. It's a smaller robotics company. They kind of play in a lot of different spaces. They had a medical division where they did software for patient self check-in and interfacing with patient medical records and these sorts of things to make it easier for you to do your billing.

Daniel Gandhi: So, for example, they have a system deployed at every VA hospital, where they have kiosks, where you can sign in yourself. You don't have to go up to a receptionist. It can check all your insurance information and you can show your bills and schedule your appointments.

Daniel Gandhi: So Vecna did things in the medical space, and they also were always trying to work in the robotic space. So they had some things that bridged those. They had a robot that was developed to, basically the medical delivery inside hospitals. So you could deliver food, you could deliver medication, you could deliver things in the same spaces that people are. And that's important when you're looking at what your medical costs look like and how much time you have. Nurses running around, just like, shuttling food back and forth. So, that was something that Vecna was trying to address through a robot like that.

[32:08] Emulating human strength in hydraulically-actuated robot arms

Daniel Gandhi: I worked on more of their research side. We were working with manipulation systems, so, robotic arms. And, what you'll see in ... When we talked earlier about industrial robotics and factory robots, you typically have a very large, heavy arm. And what people don't realize is, its lifting capability is actually really small. If we're being, I think, generous, you would say that it were like, ten to one. So, a 400 pound robot would be able to lift 40 pounds at best.

Lacey Trebaol: And obviously, not optimally.

Daniel Gandhi: Right.

Lacey Trebaol: Yeah.

Daniel Gandhi: Vecna was looking at technologies to start to flip the ratio. So, basically they used, instead of electro-mechanically actuated robotics, they hydraulically actuated the robots. And so, we had a smaller arm that might have a 40 pound weight, could lift maybe about 100 pounds. So, you have robots that can now do like, pull ups, if they wanted to.

Daniel Gandhi: The trick there is that hydraulics are far more complicated to control. You could say that hydraulics is a realm that fell out of favor at some point. They became the realm of construction equipment and heavy, industrial machinery and systems were understood for that purpose, aerospace, control surfaces. When you start getting into, want to have fine control, where I can have a robotic hand or arm that can lift objects that have substantial weight, where I want to lift a 50 pound box of something, and yet not crush the box, the control has to be far better. I was researching ways to do that, so you could actually have the same type of control that you have in electrical mechanical systems in a hydraulic system.

Lacey Trebaol: In a hydraulic system. Cool.

Daniel Gandhi: And as you start looking at robots moving into human spaces-

Lacey Trebaol: That becomes more important.

Daniel Gandhi: And doing human scale tasks ... Exactly. So, when you're talking about mobile manipulation, I want a robot that can move an environment, pick up things, put them down, reorganize things. I can't carry around a 400 pound robot that can lift 20, 30 pounds.

Lacey Trebaol: And then clear the room every time you need it to perform a task.

Daniel Gandhi: Right.

Lacey Trebaol: Okay everyone, out.

Daniel Gandhi: Right.

Steven Onzo: Thanks for listening to this episode of the Connext podcast. We hope you enjoyed it. If you have any questions or suggestions for future interviews, please be sure to reach out to us either on social media, or at podcast@rti.com. Thanks and have a great day.