Baifeng Shi
Papers
2
Total Citations
19
H-Index
2
About
Baifeng Shi is a researcher at the forefront of efficient AI and robot learning, whose work challenges conventional wisdom about model scaling. In their highly-cited 2024 paper, "When Do We Not Need Larger Vision Models?" (15 citations), Shi investigates the conditions under which smaller, more efficient vision models can match or outperform their larger counterparts—a critical insight for deploying AI in resource-constrained environments. Shi’s most impactful contribution, however, lies in robotics: their 2023 work on "Robot Learning with Sensorimotor Pre-training" introduces RPT (Robot Pre-training Transformer), a self-supervised model that learns from sequences of camera images, proprioceptive states, and actions. By masking out subsets of these sensorimotor tokens, RPT learns rich, generalizable representations without requiring expensive human annotations. This sensorimotor pre-training approach represents a paradigm shift, enabling robots to acquire foundational skills from raw experience—much like how language models learn from text. Shi’s research bridges computer vision and robotics, offering practical pathways toward more efficient, adaptable, and intelligent autonomous systems.
Research Focus
Key Achievements
Top Papers
- 1When Do We Not Need Larger Vision Models?15 citations · 2024
- 2Robot Learning with Sensorimotor Pre-training4 citations · 2023