Design and Control of a Bipedal Robotic Character

TL;DR

This work introduces a new bipedal robot, designed with a focus on character-driven mechanical features, and presents a reinforcement learning-based control architecture to robustly execute artistic motions conditioned on command signals.

Abstract

Legged robots have achieved impressive feats in dynamic locomotion in challenging unstructured terrain. However, in entertainment applications, the design and control of these robots face additional challenges in appealing to human audiences. This work aims to unify expressive, artist-directed motions and robust dynamic mobility for legged robots. To this end, we introduce a new bipedal robot, designed with a focus on character-driven mechanical features. We present a reinforcement learning-based control architecture to robustly execute artistic motions conditioned on command signals. During runtime, these command signals are generated by an animation engine which composes and blends between multiple animation sources. Finally, an intuitive operator interface enables real-time show performances with the robot. The complete system results in a believable robotic character, and paves the way for enhanced human-robot engagement in various contexts, in entertainment robotics and beyond.

Figures (12)

• Figure 1: Three instances of our robotic character performing an unscripted show. Apart from their theming, they are identical. Each robot is remote-controlled by a separate operator.
• Figure 2: Our character design and control pipeline consists of animation, mechatronic design, reinforcement learning, and run-time tools. Animation and mechatronic design form an iterative process to define the character and its motion repertoire. These inputs are used in reinforcement learning which, through imitation rewards, results in control policies that robustly execute the intended motions, conditioned on external commands. During run-time, the animation engine combines user-inputs and animation content and interfaces with the control policies. In parallel, the animation engine synchronizes show-functions with the motion.
• Figure 3: Mechanical design of our robotic character. The robot has 5 degrees of freedom per leg. The neck and head assembly has 4 degrees of freedom. The torso contains a custom communication board, a battery module, and an IMU. The onboard PC, radio receiver, and show function board are located in the head. Show functions consists of antennas, LED arrays as eyes, and a head lamp. A pair of speakers is located in the head and another pair in the bottom of the torso.
• Figure 4: Path frame illustrations for standing (top-left), and a top-view during walking (bottom-left). During standing, the path frame converges towards the average position and heading of the feet. During walking, the path frame is integrated according to the path velocity commands. The horizontal torso trajectory, shown in blue, will generally sway around the path frame to shift weight between the feet. The other images show the kinematic reference pose for the perpetual standing motion at minimum and maximum value for head commands (top) and torso commands (bottom). The input dimension from left to right are: up-down, yaw, pitch, roll.
• Figure 5: The animation engine procedurally generates the animation command, $\bm{y}_t$, based on three layers: background animation, triggered animations, and animations derived from joystick inputs. A triggered animation is blended in and out as illustrated by the green curve. In contrast, the background animation remains continuously active.