Peter So
Papers
2
Total Citations
17
H-Index
1
About
Peter So is a rising roboticist whose work is reshaping how we benchmark and represent manipulation skills. His research centers on two critical challenges: creating objective, scalable performance metrics for real-world robotics, and developing physically meaningful task representations for skill learning. In his 2024 work on the "Digital Robot Judge," So proposed a novel electronic task board system to build a task-centric performance database, enabling low-cost, repeatable evaluation of manipulation dexterity—a breakthrough that has already garnered 16 citations by addressing the long-standing difficulty of measuring real-world robotic skill gaps without expensive competitions. Complementing this, his "CITR" framework introduces a coordinate-invariant task representation, ensuring that learned manipulation skills remain robust regardless of the robot’s reference frame—a foundational step toward generalizable robotic learning. Though early in his career, So’s contributions are notable for their practical elegance: they directly tackle the reproducibility crisis in manipulation research. His work promises to accelerate progress toward robots that can match human dexterity, making him a researcher to watch in the field of robotic manipulation and benchmarking.
Research Focus
Key Achievements
Top Papers
- 1
- 2