Self-Driving on 33 Watts: How HYPR Labs Trained a Model for Just $850

Tim Kentley Klay, CEO & co-founder of HYPRLABS, joined Grayson Brulte on The Road to Autonomy podcast to discuss how the company is achieving autonomous driving in downtown San Francisco using just 33 watts of compute and zero simulation or HD maps. By prioritizing “learning velocity,” HYPR utilizes an end-to-end neural network that learns continuously from real-world driving data, avoiding the structural noise injected by classical simulation and hand-coded heuristics.

While the industry often relies on massive engineering teams and brute-force compute, HYPRLABS is executing a high-efficiency strategy with a team of just four engineers and a foundational model trained for only $850. Drawing inspiration from DeepMind’s AlphaZero, the company allows the AI to model the environment without predefined rules, using their autonomous vehicle fleet as a validation platform for a new category of robots launching next year.

Episode Chapters

• 0:00 Introduction to HYPRDRIVE
• 1:30 HYPRDRIVE
• 5:40 Learning Velocity
• 8:10 Building HYPR
• 12:23 Training the System
• 18:55 The Origins of the HYPR Approach
• 21:36 Building Trust
• 23:35 Simulation
• 27:07 $850 to Train the Model
• 30:44 HYPR Robots
• 33:22 Cameras
• 35:16 What’s Next