A Controller Design Framework For Bipedal Robots: Trajectory Optimization And Event-Based Stabilization

Abstract

This thesis presents a model-based controller design framework for bipedal robots that combines energy-efficiency with stability. We start with a physics based model for the robot and its actuators. Next, the parameters of the model are identified in a series of bench experiments. Then we formulate an energy-optimal trajectory control problem. Our energy metric is the total cost of transport (TCOT) and is defined as the energy used per unit weight per unit distance travelled. We solve the trajectory control problem using parameter optimization software and an adequately fine grid. To implement the energy-optimal solution on the physical robot, we follow a two part approach. First, we approximate the converged optimal solution with a simpler representation that sufficiently captures the optimality. The resulting walking gait is called the nominal trajectory. Second, we stabilize the nominal trajectory using an eventbased, discrete, intermittent, feed-forward controller. Our stabilizing controller tries to regulate heuristically chosen quantities in a step, like step length or step velocity, doing feedback on a few key sensor data values collected at key points in a step. Using this control framework our knee-less 2D 1 m tall 9.9 kg 4-legged bipedal robot, Ranger, achieved two feats: one, Ranger walked stably with a TCOT of 0.19, which is the lowest TCOT ever achieved by a legged robot on level terrain and, two, Ranger walked non-stop for 65 km or 40.5 miles without battery recharge or touch by a human, setting a distance record for legged robots.