QuadraTot is a robot developed at the Cornell Creative Machines Lab for use in gait learning research. The hardware was designed and 3D printed by Juan Zagal and has subsequently been used by several groups around the world. The STL files necessary to print the robot as well as all code used for research is available online (here or on our GitHub repository). A simulator is available here.
Creating gaits for legged robots is an important task to enable robots to access rugged terrain, yet designing such gaits by hand is a challenging and time-consuming process. In this project we investigate various algorithms for automating the creation of quadruped gaits. Because many robots do not have accurate simulators, we test gait learning algorithms entirely on a physical robot. We compare the performance of two classes of learning gaits: locally searching parameterized motion models and evolving artificial neural networks with the HyperNEAT generative encoding. We found that all parameter search methods outperform a manually-designed reference gait, but HyperNEAT performs better still, producing gaits nearly 9 times faster than the reference gait.
More recent work has focused on building a simulator so that a hybrid physical/simulated system can be used for designing gaits.
- Jason Yosinski ( Cornell Creative Machines Lab)
- Haocheng Shen ( Cornell Creative Machines Lab)
- Sean Lee ( Cornell Creative Machines Lab)
- Eric Gold ( Cornell Creative Machines Lab)
- Kyrre Glette ( Robotics and Intelligent Systems, University of Oslo)
- Petar Kormushev (Italian Institute of Technology)
- Jeff Clune
- Diana Hidalgo (project for Cornell's Artificial Intelligence Practicum CS 4701)
- Sarah Nguyen (project for Cornell's Artificial Intelligence Practicum CS 4701)
Many thanks to
- Jim Torresen ( Robotics and Intelligent Systems, University of Oslo)
- Hod Lipson ( Cornell Creative Machines Lab)
- Juan Zagal ( University of Chile)
Yosinski J., Clune J., Hidalgo D., Nguyen S., Zagal J., Lipson H., "Evolving Robot Gaits in Hardware: the HyperNEAT Generative Encoding Vs. Parameter Optimization", Proceedings of the European Conference on Artificial Life, 2011.
(poster) Yosinski J., Clune J., Hidalgo D., Nguyen S., Zagal J., Lipson H., "Generating Gaits for Physical Quadruped Robots: Evolved Neural Networks Vs. Local Parameterized Search", Proceedings of the Genetic and Evolutionary Computation Conference, 2011.
Haocheng Shen, Jason Yosinski, Petar Kormushev, Darwin G. Caldwell, and Hod Lipson. Learning Fast Quadruped Robot Gaits with the RL PoWER Spline Parameterization. Cybernetics and Information Technologies, Volume 12, Issue 3. 22 March 2013.
Sean Lee, Jason Yosinski, Kyrre Glette, Hod Lipson, and Jeff Clune. Evolving gaits for physical robots with the HyperNEAT generative encoding: the benefits of simulation. Applications of Evolutionary Computation, pages 540-549. 5 April 2013.
- Through the Wormhole with Morgan Freeman: Are Robots the Future of Human Evolution? Season 4, episode 7. See 7:00 - 7:45 and 9:40 - 11:10. 2013.
- Fast Company: With Evolved Brains, Robots Creep Closer To Animal-Like Learning. February 1, 2013.
- 33rd Square: As Robots Become Smarter, They Will Walk More Like Animals