Simulation of the motion after learning with our method
The motion on the real HOAP-2 robot
Coming soon!
Learning Humanoid Motion Dynamics through Sensory-Motor Mapping in a Reduced Dimensional Space
Research goal
Our research goal is to develop an algorithm that allows a humanoid robot to learn both primitive-level behaviors and high-level behaviors from human by imitation based on purely learning approach.
Introduction
Programming a humanoid robot to move is a challenging problem in robotics. Traditional approaches rely heavily on prior knowledge of the robot's physical parameters to devise sophisticated control algorithms for generating a stable gait. In this work, we provide (to our knowledge) the first demonstration that a humanoid robot can learn to walk directly by imitating a human walking gait obtained from motion capture (mocap) data. Training using human motion capture is an intuitive and flexible approach to programming a robot but direct usage of mocap data usually results in dynamically unstable motion. Furthermore, optimization using mocap data in the humanoid full-body joint-space is typically intractable. We propose a new model-free approach to tractable imitation-based learning in humanoids. We represent kinematic information from human motion capture in a low dimensional subspace and map motor commands in this low-dimensional space to sensory feedback to learn a predictive dynamic model. This model is used within an optimization framework to estimate optimal motor commands that satisfy the initial kinematic constraints as best as possible while at the same time generating dynamically stable motion. We demonstrate the viability of our approach by providing examples of dynamically stable walking in a humanoid learned from mocap data using both a dynamic simulator and a real humanoid robot.
Key idea
The key idea of this approach is learning in dimension low dimensional subspace. We use a dimension reduction algorithm such as linear PCA to construct a compact action (posture) space of a humanoid robot. The figure below shows 3D subspace of a walking gait of Fujitsu robot.
Blue points along a loop represent different robot postures during a single walking cycle. Red points mark various example poses as shown in the numbered images. The first two postures are intermediate postures between an initial stable standing pose and a point along the periodic gait loop represented by postures three through eight.
At this point, the goal is to search for a new set of postures or a new trajectory of the blue points of the figure above that gives dynamically stable motion for the robot. In the figure below, the blue points are the original data of a human walking gait from a motion capture system. The red points are the learning results which are obtained based on temporal relationship of the motor command (posture) in the low dimensional subspace and sensory feedback.