Ashok Elluswamy: Building Foundational Models for Robotics at Tesla

Ashok Elluswamy from Tesla AI team presents Tesla's mission of achieving widespread autonomous driving and humanoid robots for automating physical work. Tesla employs an end-to-end neural network using camera inputs and AI, which enhances safety by learning correct causal mappings, demonstrated by examples like navigating chickens and proactive braking. The system uses a world simulator neural network for evaluation, generating realistic videos to replay historical issues and inject new scenarios. This technology extends beyond vehicles to humanoid robots, enabling manipulation and action-based video generation. Despite challenges in evaluation and long-tail problem-solving, Tesla aims to launch CyberCabs and Optimus Robots to reduce transportation costs and improve global productivity. The presentation concludes with a Q&A, addressing the camera-only approach, the use of rewards in the simulator, and future voice command integration.