A reinforcement learning based neural network architecture for obstacle avoidance in multi-fingered grasp synthesis

The ability to learn from interaction with the exterior world as well as variability are two main features of living organisms. The aim of this study is to present and discuss the property of a stochastic reinforcement learning-based model of upper limb posture generation that exhibits both properties. The capacity of the model to discover suitable postures satisfying task and obstacle avoidance constraints is demonstrated by simulation. Also, task equivalent configurations that can be linked to recent findings in the motor control literature are generated by the proposed formalism due to its stochastic nature.