Model-based control through numerical optimization
We are developing new methods for control optimization aimed at real-time robotic control. This is a collaboration betwen Zoran Popovic's group, which developed high-fidelity controllers for physics-based animation, and Emo Todorov's group, which developed efficient algorithms for optimal control as well as a new physics engine tailored to control optimization. Our approach combines offline learning of semi-global value functions and control policies with online trajectory optimization or model-predictive control (MPC). We recently used MPC to make a 3D robot bounce two ping-pong balls on the same paddle and to generate complex locomotion behaviors, including swimming, hopping, and getting up in real-time. We are now extending this work to humanoid robots by increasing the efficiency of our optimization algorithms as well as porting our physics engine to GPUs.
People
- Tom Erez
- Paul Kulchenko
- Igor Mordatch
- Zoran Popović
- Yuval Tassa
- Emanuel Todorov
Publications
- A convex, smooth and invertible contact model for trajectory optimization (2011)
- Complex object manipulation with hierarchical optimal control (2011)
- First-exit model predictive control of fast discontinuous dynamics: Application to ball bouncing (2011)
- Infinite-horizon model predictive control for nonlinear periodic tasks with contacts (2011)
- Optimal limit-cycle control recast as Bayesian inference (2011)
- Policy gradient methods with model predictive control applied to ball bouncing (2011)
- Implicit nonlinear complementarity: A new approach to contact dynamics (2010)
- Stochastic complementarity for local control of discontinuous dynamics (2010)

cs.