Catalyst with Shayle Kann Cracking the code on autonomous trucking
May 7, 2026
Eyal Cohen, founder and CEO of Humble Robotics and former Apple and Uber autonomy lead, discusses building cabless electric autonomous trucks from the ground up. He explains why freight autonomy differs from passenger cars. He covers camera-first perception and visual language models, the hub-to-hub business model, electrification and charging, and regulatory challenges for cabless trucks.
AI Snips
Chapters
Transcript
Episode notes
Why Highways Are Harder For Autonomy Than They Look
- Autonomous trucking felt easier than city passenger autonomy in 2016 but proved harder due to high-speed, heavy-vehicle edge cases.
- Trucks encounter rarer but higher-consequence events, long stopping distances, and cannot safely 'just stop' like a passenger car, making highways deceptively difficult.
Vision Language Models Shift The Sensor Balance
- The autonomy tech stack has shifted from handcrafted sensors and HD maps toward machine-learned perception dominated by vision-language models (VLMs).
- Eyal says Humble is camera-first now because open VLMs provide rich image understanding previously unavailable without heavy LiDAR reliance.
Use All Sensors To Minimize Risk Not To Win Debates
- Prioritize multimodal sensing for truck autonomy to maximize safety rather than adopt a dogmatic camera-or-LiDAR stance.
- Use camera for lights/semantics, LiDAR for long-range/low-light, and radar for complementary robustness on heavy vehicles.