Max Robotics — Global Service Robot Directory + US Certification

About

Kanishka Rao is a pioneering robotics and machine learning researcher whose work sits at the intersection of large language models, computer vision, and real-world robotic control. Best known for contributions to Google's landmark RT-1 and RT-2 robotics transformer frameworks — together accumulating over 800 citations — Rao has helped establish the paradigm of training generalizable robotic policies from large-scale, diverse datasets. His influential work on "Do As I Can, Not As I Say" (516 citations) demonstrated how language models can be grounded in physical robotic affordances, enabling robots to interpret and execute complex natural language instructions meaningfully rather than superficially. Rao has also made significant contributions to the sim-to-real transfer problem, developing RL-CycleGAN (154 citations) and RetinaGAN to bridge the visual gap between simulation and physical environments — a critical bottleneck in scalable robot learning. His more recent work explores open-vocabulary scene representations, code-as-policies prompting strategies, and scalable offline reinforcement learning via Q-Transformer, reflecting a broad and forward-looking research vision. Across his portfolio, Rao consistently advances the goal of building robots that can generalize across tasks, environments, and instructions — bringing truly capable robotic systems meaningfully closer to real-world deployment.