The Self-Driving Startup Nobody Saw Coming | E2289

Wave's CEO Alex Kendall returned to discuss how the company has advanced its contrarian technical and commercial strategy for bringing scalable self-driving systems to market. The conversation centers on Wave's use of end-to-end learned world models, its decision to license rather than own hardware or operate fleets, and the shift from scientific risk to engineering and deployment execution.

Technical Foundations and Model Development

Wave built world models years before they became standard in robotics. These models serve dual purposes: they create rich internal representations of driving scenes by learning to predict future states from actions, and they function as controllable simulators for training and validation. The approach favors generalization across environments over optimization in narrow domains, which Kendall credits for the company's ability to handle diverse sensor suites, weather, and geographies without bespoke hardware.

Between successive generations, Wave scaled parameter counts, data volume, and algorithmic controls. The models now ingest camera, radar, and lidar inputs from multiple partners and support promptable simulation for safety-critical edge cases. Kendall notes that this unified stack removes the need to maintain separate systems for different autonomy levels or vehicle types.