Learning Biped Locomotion

[source] Robotics & Automation

[year] 2007

 

They improved biped walking controllers based on an approximated Poincare map using a model-based RL framework.

 

model-based RL algorithm for biped walking in which robot learns to appropriately modulate an observed walking pattern.

minimum jerk criterion检测中via-point

RL控制via-point,是基于庞映射的

是从single support phase and the controlled via-points 到 next single support phase的映射

 

 

1.在每个via-point (pi/2 及 3pi/2)处定义了状态， 此时可采取action

2.对via-points之间的用PDs完成

3.决定下一via-point的映射需要学习得到

4.用Guassian distribution来表示policy