Tesla’s Dojo Supercomputer: Revolutionizing Self-Driving with 4D Processing

Tesla is building a supercomputer called Dojo to process vast amounts of video data, aiming for L5 self-driving capabilities through a 4D system.

Elon Musk recently announced Tesla’s recruitment drive for AI and chip talent to work on the Dojo supercomputer project, which aims to process an unprecedented amount of video data for neural network training. Musk emphasized that the full self-driving (FSD) improvement will be a quantum leap due to a fundamental architectural rewrite, not an incremental tweak.

The Dojo project reflects Musk’s ambition to achieve L5 autonomy by upgrading Tesla’s autopilot system from a 2.5D to a 4D infrastructure. This transition involves incorporating temporal information, enabling the system to predict intentions and potential interactions with objects, akin to human perception.

UC Berkeley researcher Dr. Fisher Yu explains that moving from 2.5D to 4D representations allows for more robust autonomous driving systems, capable of leveraging 3D information for safer routes and predicting vehicle functionality. The integration of temporal data from video feeds enhances the system’s ability to make accurate predictions and decisions.

Tesla’s recent patent on generating ground truth for machine learning from time series elements further elucidates Musk’s vision for a 4D system, which involves capturing accurate image data and odometry data over time to improve the precision of autonomous driving.

Andrej Karpathy, Tesla’s Senior AI Director, highlighted the goal of the Dojo supercomputer to significantly increase performance at a lower cost. If successful, the Dojo supercomputer and the 4D Autopilot system could give Tesla a substantial lead in the race to achieve L5 self-driving capabilities.

Musk has hinted at a limited public release of the 4D Autopilot FSD upgrades within 6 to 10 weeks. For those interested in contributing to this groundbreaking technology, Tesla is hiring engineers in Palo Alto, Austin, and Seattle, with the possibility of remote work for exceptional candidates.