Perry Dong

Papers

Total Citations

H-Index

About

Perry Dong is a rising researcher at the intersection of reinforcement learning (RL), imitation learning, and robotics, whose work is reshaping how autonomous systems acquire complex skills. His primary contributions lie in bridging the gap between offline RL and interactive imitation learning, making data-driven skill acquisition more practical for real-world robotic applications. In his influential 2023 paper "RLIF: Interactive Imitation Learning as Reinforcement Learning" (7 citations), Dong demonstrated how interactive imitation learning can be reframed as an RL problem, offering a more accessible alternative to traditional RL for domains like robotics. His companion work, "Action-Quantized Offline Reinforcement Learning for Robotic Skill Learning" (4 citations), tackles the critical challenge of distributional shift in offline RL by introducing action quantization, enabling policies derived from static datasets to outperform their data-collecting counterparts. These contributions are particularly notable for their focus on practical deployment—addressing the real-world constraints of robotics where online exploration is costly or dangerous. Dong’s research is gaining traction for its elegant synthesis of imitation learning’s convenience with RL’s optimality guarantees, positioning him as a promising voice in the push toward more sample-efficient, deployable robot learning systems.