Chris Urmson on the Evolution, Challenges, and Future of Self-Driving Cars from DARPA to Aurora

Chris Urmson, a pioneer in autonomous vehicles, discusses the journey from the DARPA Grand and Urban Challenges to the current state of self-driving technology at Aurora. He emphasizes that the initial belief in the possibility of autonomous driving, despite its perceived difficulty, was a crucial first step. The evolution involved significant technological leaps, including high-definition (HD) mapping and multi-beam lidar, which enabled vehicles to understand and navigate complex environments. Urmson highlights the continuous challenge of dealing with unpredictable actors and the vast scale of real-world urban environments compared to controlled test tracks.

A significant distinction Urmson makes is between driver-assistance systems (Level 2) and truly self-driving vehicles (Level 4/5). He expresses strong concerns about Level 2 systems due to the human factor of over-trust and complacency, arguing that people will inevitably misuse them, leading to dangerous situations. He believes the technological and economic paths for these two levels are divergent, with Level 2 focusing on cost reduction for partial assistance, while Level 4/5 requires robust, high-performance systems for full autonomy, prioritizing safety above all else. He also differentiates between the "cheapest" sensor suite and an "economically viable" one that actually works safely.

Urmson advocates for a comprehensive sensor suite, including lidar, cameras, and radar, to achieve robust autonomy, countering the notion that lidar is a "crutch." He argues that any technology that accelerates the deployment of safe self-driving cars should be utilized, given the high number of road fatalities. For demonstrating safety, he suggests a multi-faceted approach involving diligent functional safety processes, extensive testing (simulation, unit, decomposition), and on-road data. He dismisses simple metrics like "disengagements" as easily manipulated and proposes evaluating performance against human failure rates for specific driving tasks.

The conversation touches on the philosophical implications of entrusting one's life to a machine, acknowledging public skepticism as reasonable. Urmson believes that direct experience will ultimately win over public trust, as the technology becomes mundane and its benefits in terms of safety, mobility access, and cost reduction become evident. The discussion underscores the immense engineering and societal challenges in bringing autonomous vehicles to scale, emphasizing the need for rigorous development, clear communication, and collaboration with regulatory bodies to ensure public safety and acceptance.

"the high order bit was that it could be done"

"I think there's a certain benefit to naivete right that if you don't know how hard something really is you you try different things"

"I think the other big one though is to see people for who they can be not who they are"

"the real technology that unlocked that was HD mapping"

"the biggest thing is that the you know the the actors are truly unpredictable"

"I think it is undeniable that people walk around without you know lasers in their forehead and they can get into vehicles and drive them"

"any technology that we can bring to bear that accelerates the this techno you know self-driving technology coming to market and saving lives is technology we should be using"

"I still believe several things around this one is people will over trust the technology"

"I don't think so if people truly understood the risks and internalized it then then sure you could do that safely but that that's a world that doesn't exist"

"it becomes mundane which is which is exactly what you want a technology like this to be"