Off-line path integral reinforcement learning using stochastic robot dynamics approximated by sparse pseudo-input Gaussian processes: Application to humanoid robot motor learning in the real environment.
Norikazu SugimotoJun MorimotoPublished in: ICRA (2013)
Keyphrases
- humanoid robot
- gaussian processes
- real robot
- motor learning
- real environment
- fully autonomous
- reinforcement learning
- biologically inspired
- motion planning
- gaussian process
- multi modal
- motor skills
- mobile robot
- human brain
- augmented reality
- covariance function
- machine learning
- learning process
- high fidelity
- active learning
- gaussian process regression
- model selection
- learning experience
- dynamic environments