This paper presents a framework to generate energy efficient dynamic human-like walk for a Nao humanoid robot. We first extend the inverted pendulum model with the goal of finding an energy efficient and stable walking gait. In this model, we propose a leg control policy which utilizes joint stiffness control. We use policy gradient reinforcement learning to identify the optimal parameters of the new gait for a Nao humanoid robot. We successfully test the control policy in a simulator and on a real Nao robot. The test results show that the new control policy realizes a dynamic walk that is more energy efficient than the standard walk of Nao robot.
|Title of host publication||2014 IEEE International Conference on Autonomous Robot Systems and Competitions, ICARSC 2014|
|Publisher||IEEE Computer Society|
|Number of pages||6|
|Publication status||Published - 2014|
- Humanoid Robot
- Learning Control