The Next Byte Podcast · 23 May 2023 · 20m
Hosts Daniel and Forbode discuss research from the University of Michigan that uses mixed reality to accelerate autonomous vehicle (AV) testing. By superimposing virtual 'bad driver' vehicles onto real physical cars on the MCity test track, researchers can simulate rare safety-critical events without real-world risk. This approach, leveraging data from smart intersections in Ann Arbor and Detroit, reduces the required testing miles by 99.99%, addressing the 'curse of rarity' in AV training data. The hosts highlight how this method lowers costs and time while improving safety for companies like Waymo, Tesla, and Cruise.
Where this conversation describes AI being used to build or deliver a product. Part of the AI in practice tracker.
Engineering & simulationdense reinforcement learning99.99% reduction in testing miles
Researchers used dense reinforcement learning to train virtual background vehicles to drive like bad drivers, creating rare safety-critical scenarios for testing autonomous vehicle algorithms without real-world risk.
“So basically what they're doing they're using dense reinforcement learning. to teach these, you know, these virtual background cars to drive like bad drivers.”
Speaker 2 · Hear it at 10:18
early-stage companies
Other shows on the same ground, closest first.
Mo ElShenawy, Senior VP of Engineering at Cruise (a GM subsidiary), discusses the engineering challenges of developing fully autonomous vehicles (L4/L5). He explains why Cruise targets San Francisco's complex environment to accelerate learning and details the hardware stack, including LiDAR, radar, and cameras, fused for perception and prediction. ElShenawy emphasizes that because state-of-the-art literature cannot yet solve L4/L5 driving, Cruise is inventing new algorithms and building a massive simulation and data infrastructure to ensure safety and scale.
Shimon Whiteson, a Professor at the University of Oxford and Senior Staff Research Scientist at Waymo, explains how deep reinforcement learning and imitation learning are used to make driverless vehicles smarter. He details the acquisition of his spinout company, Latent Logic, by Waymo in 2019 to improve simulation realism by modeling human road users. Whiteson discusses Waymo's multimodal sensor architecture (cameras, lidar, radar), the importance of data curation for rare safety-related conflicts, and the company's patient, capital-intensive strategy that has led to its commercial robotaxi leadership.
Matt Puchalski, founder and CEO of Bucket Robotics, explains how his Y Combinator-backed startup uses synthetic data generated from STEP files to automate visual inspection in manufacturing, bypassing the need for physical 'golden samples.' He contrasts this approach with legacy systems from Keyence and Cognex, highlighting how starting from CAD models allows for rapid simulation of good and bad parts. Puchalski also shares his background in autonomous vehicles at Argo AI and Latitude AI, discussing the unique reliability challenges of deploying Level 4 AVs and how those lessons inform his current work in flexible manufacturing.
Vassia Simaiaki, Head of AI Vision and Graphics at Wayve, explains the company's end-to-end AI approach to autonomous driving, which relies on camera data rather than expensive hardware like LiDAR or HD maps. She details how Wayve uses novel view synthesis and a neural simulator called NeuroSim to generate synthetic data, allowing their models to generalize to unseen scenarios like the 'plastic bag problem' or complex London traffic. The discussion highlights Wayve's partnership with Asda to expand grocery delivery coverage and their focus on proving safety through simulation to build public trust.
Semicast research analyst Colin Barden argues that the autonomous vehicle (AV) industry's promise of saving lives is a hubristic distraction from the more viable path of advanced driver assistance systems (ADAS) and driver monitoring. He contrasts the 'left fork' of tech companies like Waymo with the 'right fork' of traditional automakers focusing on collaborative driving. The episode then shifts to a roundtable with EE Times editors Junko Yoshida, Sally Ward-Foxton, and Anne-Françoise Palais on the surge of voice recognition technology in consumer electronics, driven by the need for edge processing to reduce cloud costs and improve privacy.
The summary, chapters and topics are generated from the episode’s audio by this directory, not written by The Next Byte Podcast. The episode itself belongs to its publisher.