Rishabh Garg is co-founder and chief technology officer of Gradient Robotics, which is building a robotic workforce for data-center construction. He specializes in the difficult boundary between intelligent control policies and the hardware that executes them: sensors, actuators, operating-system schedules, and communication networks must stay synchronized for robots to move reliably.
Garg previously worked on robotics and embedded systems at Amazon and Tesla, and was affiliated with Tesla Optimus in 2025. He subsequently co-founded Gradient Robotics with Jingxiang Mo, applying his experience in software-hardware integration to the physical infrastructure supporting modern computing.
What makes robotic systems reliable
- Control failures can originate in communications. A hesitant or jittery robot may have a sound policy but delayed commands, stale sensor readings, or desynchronized threads.
- CAN bus throughput constrains control-loop performance. Sensor and actuator traffic can consume a substantial portion of each control cycle; pipelining transmit, receive, and policy execution improves throughput but introduces synchronization risks.
- Cycle-time analysis reveals hidden timing failures. An external CAN transceiver, timestamped
candump traffic, and message-interval plots can expose queued commands, delayed transmission, and stale observations that otherwise resemble model failures.
- Logging and scheduling affect physical behavior. Disk writes can stall control loops, microcontroller logging can compound packet loss, and aggressive process priorities can starve kernel pathways delivering sensor data. Condition variables, semaphores, timing buffers, and carefully chosen priorities help protect execution.
His AI Engineer robotics presentation frames dependable physical intelligence as an end-to-end robotics systems architecture problem, extending from policy execution through communication and actuation.
