Motor Babble: Morphology-Driven Coordinated Control of Articulated Characters Avinash Ranganath Clemson University Clemson, SC, USA arangan@clemson.edu Avishek Biswas Clemson University Clemson, SC, USA avisheb@clemson.edu Ioannis Karamouzas Clemson University Clemson, SC, USA ioannis@clemson.edu Victor B. Zordan Clemson University Clemson, SC, USA vbz@clemson.edu ABSTRACT Locomotion in humans and animals is highly coordinated, with many joints moving together. Learning similar coordinated loco- motion in articulated virtual characters, in the absence of reference motion data, is a challenging task due to the high number of degrees of freedom and the redundancy that comes with it. In this paper, we present a method for learning locomotion for virtual charac- ters in a low dimensional latent space which defnes how diferent joints move together. We introduce a technique called motor babble, wherein a character interacts with its environment by actuating its joints through uncoordinated, low-level (motor) excitations, re- sulting in a corpus of motion data from which a manifold latent space is extracted. Dimensions of the extracted manifold defne a wide variety of synergies pertaining to the character and, through reinforcement learning, we train the character to learn locomotion in the latent space by selecting a small set of appropriate latent dimensions, along with learning the corresponding policy. CCS CONCEPTS · Computing methodologies → Physical simulation; Learning latent representations; Reinforcement learning. KEYWORDS character animation, physics-based control, reinforcement learning, animal locomotion ACM Reference Format: Avinash Ranganath, Avishek Biswas, Ioannis Karamouzas, and Victor B. Zordan. 2021. Motor Babble: Morphology-Driven Coordinated Control of Articulated Characters. In Motion, Interaction and Games (MIG ’21), No- vember 10ś12, 2021, Virtual Event, Switzerland. ACM, New York, NY, USA, 10 pages. https://doi.org/10.1145/3487983.3488291 Permission to make digital or hard copies of all or part of this work for personal or classroom use is granted without fee provided that copies are not made or distributed for proft or commercial advantage and that copies bear this notice and the full citation on the frst page. Copyrights for components of this work owned by others than ACM must be honored. Abstracting with credit is permitted. To copy otherwise, or republish, to post on servers or to redistribute to lists, requires prior specifc permission and/or a fee. Request permissions from permissions@acm.org. MIG ’21, November 10ś12, 2021, Virtual Event, Switzerland © 2021 Association for Computing Machinery. ACM ISBN 978-1-4503-9131-3/21/11. . . $15.00 https://doi.org/10.1145/3487983.3488291 Figure 1: Locomotion learned from morphologically specifc motor babble. 1 INTRODUCTION Despite recent advances in trajectory optimization and reinforce- ment learning, it remains challenging to learn motor skills for physics-based articulated characters. While human motion data has been used to bootstrap control for humanoid characters, animating complex non-human characters like those seen in Figure 1 presents a challenging control problem which can be under specifed and prohibitively high dimensional. While there is typically an ample space of control policies to accomplish motor tasks, not all results lead to natural and coordinated motion. This paper introduces an approach that attempts to mitigate this problem by extracting coor- dinated motor activations which are drawn from the character’s own dynamics directly using a technique we call łmotor babblež after its inspiration taken from robotics. State-of-the art deep reinforcement learning (DRL) approaches excel at generating natural control policies for physically simulated humanoids, and, more recently, for simple quadrupeds by imitating motion capture clips of expert behaviors [Park et al. 2019; Peng