Xinyang Geng
Papers
7
Total Citations
180
H-Index
6
About
Xinyang Geng is a leading researcher at the intersection of reinforcement learning (RL) and robotics, with a focus on enabling complex, real-world manipulation and control. His work spans tensegrity robot locomotion, multi-stage manipulation, and offline model-based optimization. Geng made early, impactful contributions to controlling tensegrity robots—lightweight, cable-actuated structures ideal for planetary exploration—demonstrating that deep RL could master their complex dynamics (92 citations). He later advanced the field of robotic manipulation by tackling the notoriously difficult problem of cable routing, using hierarchical imitation learning to break down multi-stage tasks (46 citations). Geng has also been instrumental in developing offline RL and optimization methods, including conservative objective models for data-driven design (10 citations) and benchmarks like Design-Bench (8 citations), which standardize evaluation for offline model-based optimization. His recent work on action-quantized offline RL for robotic skill learning (4 citations) further pushes the boundaries of learning from static datasets. Through these contributions, Geng has established himself as a key figure in making RL practical for complex, real-world robotic systems.
Research Focus
Key Achievements
Top Papers
- 1Deep reinforcement learning for tensegrity robot locomotion92 citations · 2017
- 2Multistage Cable Routing Through Hierarchical Imitation Learning46 citations · 2024
- 3Deep Reinforcement Learning for Tensegrity Robot Locomotion11 citations · 2016
- 4Conservative Objective Models for Effective Offline Model-Based Optimization10 citations · 2021
- 5
- 6
- 7