At the Speed of Light: Next-Gen Vehicle Development with DDS and Lidar
Written by Neil Puthuff & Nick Mishkin
January 18, 2022
A few months ago, the first-ever autonomous racing competition using full-size Indy Lights race cars was held at the Indianapolis Motor Speedway. The original Indy Autonomous Challenge race was a culmination of more than 18 months of increasingly difficult challenges for the university teams, who achieved speeds of more than 150 mph (241 kmh) at the famed oval.
RTI and Luminar Technologies were both sponsors for these challenges: Luminar, an automotive technology company, provided their outstanding Hydra H3 lidar sensors and support, while RTI provided its Connext® software data communications framework. Both played key roles in securing victory and the $1 million grand prize for the winning team: Technische Universität München (TUM).
With their complementary technologies, RTI and Luminar worked together to achieve racing-grade perception with ultra-low latency for capturing real-world event data from the race track and delivering it to the racing software. Since this unique technology pairing played a part in TUM’s historic IAC victory, let’s dive into how it all came about.
First of all, it begins with the lidar sensor. Many lidar manufacturers today employ a single 360-degree rotating sensor head to collect data. This means that it’s looking ‘forward’ only ⅓ of the time or less and effectively spending much more time and energy to get data points on areas that don't matter. At racing speeds, having the highest resolution where you need it is the name of the game.
Luminar Lidar is different and is uniquely suited for high-speed racing: instead of a single 360-degree spinning sensor, the IAC vehicles were equipped with three (3) Luminar sensors, each covering a 120-degree field of view with the capability to focus higher resolution only in the areas that matter most -- particularly at the horizon, 200+ meters away. By having the greatest amount of points on the horizon and targets at range, the IAC vehicles were able to anticipate the curves of the track and any obstacles early enough to avoid said obstacles and reach average speeds of 139mph.
Secondly, the other piece of the puzzle is the communications software. To say there is a lot of data flying around in autonomous driving is an understatement of epic proportions. And for the IAC vehicles, it was critical to capture the lidar data in real-time, in order to literally keep the car on track. So to help various participants in the IAC race, RTI wrote the device driver software for the Luminar lidar and other sensors used in the IAC vehicles, which was implemented as a set of ROS 2 device drivers. But that was merely phase one. Due to the enhanced performance capability of the Luminar lidar and its importance to the autonomous racing software, something more was needed.
To achieve the lowest latency as is needed in this high-speed environment, RTI used the approach outlined in an earlier RTI blog to create a native-Connext device driver for the Luminar lidar. A considerable amount of latency and system resources are consumed within ROS 2 itself, but because ROS 2 is implemented on the DDS framework, this overhead can be bypassed. As shown in that blog, it’s very straightforward to convert a ROS 2 application to native Connext, resulting in extremely low latency and full access to the capabilities of Connext, while still maintaining full interoperability with ROS 2.
How much of a difference does this approach make? Frankly, the improvement is astounding: the optimized driver had less than half of the latency of the ROS 2 driver, and consumed less than a third of its system overhead. It works with multiple lidar sensors and provides the management capability to tune the dataflows for peak system performance. And because it maintains compatibility with ROS 2, it was a drop-in replacement for the three (3) instances of the ROS 2 lidar driver. No modifications were needed to the rest of the system software.
This enhanced lidar driver was used by the winning team, TUM, to produce a flawless performance at lap speeds averaging over 135 mph (217 kmh), despite having less-than-ideal racing conditions (54F/12C and cloudy) and the loss of an entire practice day due to heavy rainfall on the day before the race.
These poor conditions were a factor in a nerve-rattling spinout of the TUM vehicle that happened just days before the race (you can see the plumes of tire smoke in the Lidar display of this replay). Clearly, the teams were pushing the vehicles to their limits on the track!
Racing has long been a crucible for new technologies, from the humble rear-view mirror to the latest in autonomous driving software. RTI and Luminar are honored to have taken part in that tradition at the Indy Autonomous Challenge– both the race we described here, and the follow up race held in Las Vegas on January 7.
If you need this level of race-winning performance in your Automotive/Autonomous or ROS 2 system(s), please contact us at firstname.lastname@example.org.
About the authors:
Neil Puthuff is a Senior Software Integration Engineer for Real-Time Innovations with a focus on Robotics and Automotive Systems, and RTI team lead for the Indy Autonomous Challenge. Versed in hardware as well as software, he created the processor probes and replay debugging products at Green Hills Software before joining RTI. Neil is a named inventor on more than a dozen US patents.
Nick Mishkin, Business Development, Luminar.