Episode Summary
Executive Summary: This episode argues that autonomous trucking has lagged passenger AVs not because highways are simpler, but because trucks face harder edge cases, higher safety stakes, and worse economics. Ayal Cohen explains why Humble Robotics is taking a clean-sheet, cabless, electric approach built around modern vision-language models to reduce cost, improve safety, and speed deployment.
Main Topics: Why trucking autonomy has lagged passenger AVs (Priority: 5/5): Cohen explains that early autonomous trucking teams assumed highways would be easier than dense urban streets, but higher speeds, longer stopping distances, rare-but-severe edge cases, and the inability to safely stop make trucking harder in practice. The shift from legacy autonomy stacks to VLM-era AI (Priority: 5/5): He contrasts older HD-map, hand-coded, LiDAR-heavy systems with today’s camera-first, VLM-enabled approach that provides richer scene understanding and reduces the amount of custom feature engineering required. Humble Robotics’ clean-sheet cabless vehicle design (Priority: 5/5): Humble is building an autonomous electric Class 8 vehicle from scratch, without a cab, treating the whole vehicle as smart rather than retrofitting an existing tractor. Cohen says this improves visibility, safety, and long-term unit economics. Sensor strategy: multi-modal safety over dogma (Priority: 4/5): Cohen rejects the camera-only versus LiDAR-only debate and argues that driverless trucking today should use camera, LiDAR, and radar together, with different sensors handling different tasks and environmental conditions. Electric trucking and operational fit (Priority: 4/5): The company’s platform is electric because fully autonomous freight is easier to automate with EV charging than diesel refueling, and because electric trucks make the most sense in short-haul, port, depot, and local freight use cases. Rollout path, economics, and regulation (Priority: 5/5): Cohen expects long-haul autonomous trucks to appear first in constrained hub-to-hub deployments, likely in states like Texas. He emphasizes that trucking must win on cost, while regulators are increasingly enabling testing and deployment.
Key Arguments: Passenger AVs reached commercialization first because the urban problem, while difficult, is more manageable than high-speed trucking edge cases and safety constraints. Highway autonomy is harder than it looks: trucks travel fast, are extremely heavy, have long stopping distances, and cannot safely fail by simply pulling over or stopping. Legacy autonomy companies had to continually re-engineer stacks; a new entrant can start with modern VLMs and camera-first systems from day one. Humans effectively drive with vision, so VLMs bring a more natural, information-rich perception layer that can accelerate autonomy development. A cabless design is not just a simplification; it removes cost and weight, improves rear visibility, and makes full-vehicle autonomy more coherent. For trucking, multi-sensor redundancy matters more than ideological purity: camera, LiDAR, and radar should be combined to maximize safety. Autonomous trucking adoption will depend on unit economics, not novelty; unlike consumer ride-hailing, freight buyers will not pay a premium for coolness. Electricity and autonomy reinforce each other because fully hands-off freight is far easier to automate with electric charging than with diesel refueling. Regulatory progress is uneven but moving: California recently became permissive for driverless trucking under conditions, while Texas remains favorable for testing and early deployment.
Data Points: Waymo public ride service launch: December 2018 - Used as a benchmark for how far passenger autonomy has advanced relative to trucking Waymo cumulative public-road miles: 200 million miles - Cited to illustrate the maturity of passenger autonomous driving U.S. freight moved by trucks: 11.2 billion tons last year - Used to show the scale of the trucking market Autonomous trucking industry start: 2016 - Cohen says serious non-defense autonomous trucking efforts began around this time Autonomous trucking pioneers: 2 companies: Starsky Robotics and Otto - Named as the first serious industry players Truck weight at full gross weight: Up to 80,000 pounds - Illustrates why stopping distance and safety margins are critical Highway test distance cited: 200 miles - Cohen says a truck could likely travel this far on a highway with minimal perception if it stayed in lane and speed LiDAR range discussed: 300-400 meters - Described as the long-range LiDAR target needed to handle truck stopping distance Driverless run example: A commercial driverless run reported by Ba Auto - Mentioned as a recent milestone in the space Truck operating cost: About $2.30-$2.40 per mile - Industry-guide estimate referenced for long-haul trucking economics Driver wage component: About $1,100 per mile - Cohen cites this as the labor portion of trucking cost, roughly 30%-40% of total cost Tractor price: $150,000-$250,000 - Typical price range for a conventional tractor Electric truck price: $400,000-$500,000 - Illustrates the current premium on electric trucking hardware VPP customer devices: 2.5 million devices - From the ad read on EnergyHub, not core to the interview but present in the transcript VPP dispatchable capacity: 3.4 gigawatts - From the ad read on EnergyHub, not core to the interview but present in the transcript
Pivotal Quotes: "The highways were way more difficult than we expected. They're challenging." — Ayal Cohen: Explaining why autonomous trucking has advanced more slowly than passenger AVs "The way I think about it is what's the best technology for the moment that makes us the safest." — Ayal Cohen: Describing Humble’s non-dogmatic multi-sensor strategy "If you remove a cab, you've taken out some significant cost from the vehicle. You've also taken out significant weight from the vehicle." — Ayal Cohen: Explaining the rationale for Humble’s cabless clean-sheet truck
Implications: Autonomous trucking is likely to emerge first in constrained, high-value routes, not everywhere at once. The winners will pair modern AI perception with safety, lower vehicle cost, and workable charging and regulatory pathways.