Adaptive Tensegrity Locomotion on Rough Terrain via Reinforcement Learning

This project is a classic academic research artifact associated with a 2018 paper. While the application of Guided Policy Search (GPS) to tensegrity robots was a novel combination at the time, the project lacks any modern momentum. With 0 stars and no activity in several years, it serves purely as a historical reference implementation. Tensegrity robotics is a highly niche field—primarily of interest to NASA for planetary exploration—meaning frontier labs are unlikely to compete here. However, the RL techniques used (GPS) have been largely superseded by more robust and sample-efficient algorithms like PPO, SAC, and modern Sim-to-Real frameworks. The defensibility is minimal because the code is dated, the community is non-existent, and the specific robotics hardware (6-bar tensegrity) has limited commercial application. Any modern team tackling this problem would likely start with a contemporary stack rather than building on this repository.