Max Robotics — Global Service Robot Directory + US Certification

About

Stephen Tian is a robotics researcher whose work spans tactile sensing, robot learning, video prediction, and the application of foundation models to embodied intelligence. His early landmark contribution, "Manipulation by Feel: Touch-Based Control with Deep Predictive Models" (121 citations), demonstrated how deep learning could harness tactile feedback for continuous, non-prehensile robotic manipulation — a notoriously difficult problem that traditional physics-based approaches struggled to solve. This work was complemented by OmniTact, a novel multi-directional, high-resolution touch sensor that expanded the sensing capabilities available to robotic systems. Tian's research has increasingly focused on scalable robot learning. He contributed to RoboNet and the large-scale DROID dataset (108 citations), advancing the field's understanding of how diverse, real-world manipulation data can improve generalization. His MaskViT work (45 citations) showed that masked visual pre-training enables transformers to build powerful video prediction models for planning. Most recently, his survey on foundation models in robotics (163 citations) has become a defining reference for the field, synthesizing how internet-scale pretrained models can overcome the limited adaptability of task-specific robot learning systems. Across these contributions, Tian's research consistently bridges perception, learning, and scalable data collection to push robotic manipulation toward greater generality and robustness.