Teleoperated robotic characters can perform expressive interactions with humans, relying on the operators’ experience and social intuition. In this work, we propose to create autonomous interactive robots, by training a model to imitate operator data. Our model is trained on a dataset of human-robot interactions, where an expert operator is asked to vary the interactions and mood of the robot, while the operator commands as well as the pose of the human and robot are recorded. Our approach learns to predict continuous operator commands through a diffusion process and discrete commands through a classifier, all unified within a single transformer architecture. We evaluate the resulting model in simulation and with a user study on the real system. We show that our method enables simple autonomous human-robot interactions that are comparable to the expert-operator baseline, and that users can recognize the different robot moods as generated by our model. Finally, we demonstrate a zero-shot transfer of our model onto a different robotic platform with the same operator interface.

Procedural animation has seen widespread use in the design of expressive walking gaits for virtual characters. While similar tools could breathe life into robotic characters, existing techniques are largely unaware of the kinematic and dynamic constraints imposed by physical robots. In this paper, we propose a system for the artist-directed authoring of stylized bipedal walking gaits, tailored for execution on robotic characters. The artist interfaces with an interactive editing tool that generates the desired character motion in realtime, either on the physical or simulated robot, using a model-based control stack. Each walking style is encoded as a set of sample parameters which are translated into whole-body reference trajectories using the proposed procedural animation technique. In order to generalize the stylized gait over a continuous range of input velocities, we employ a phase-space blending strategy that interpolates a set of example walk cycles authored by the animator while preserving contact constraints. To demonstrate the utility of our approach, we animate gaits for a custom, free-walking robotic character, and show, with two additional in-simulation examples, how our procedural animation technique generalizes to bipeds with different degrees of freedom, proportions, and mass distributions.

This paper presents a compliant locomotion framework for torque-controlled humanoids using model-based whole-body control. In order to stabilize the centroidal dynamics during locomotion, we compute linear momentum rate of change objectives using a novel time-varying controller for the Divergent Component of Motion (DCM). Task-space objectives, including the desired momentum rate of change, are tracked using an efficient quadratic program formulation that computes optimal joint torque setpoints given frictional contact constraints and joint position / torque limits. In order to validate the effectiveness of the proposed approach, we demonstrate push recovery and compliant walking using THOR, a 34 DOF humanoid with series elastic actuation. We discuss details leading to the successful implementation of optimization-based whole-body control on our hardware platform, including the design of a “simple” joint impedance controller that introduces inner-loop velocity feedback into the actuator force controller.

This paper provides an overview of the embedded joint-space control approach developed for THOR, a new series elastic humanoid. The 60 kg robot features electromechanical linear series elastic actuators (SEAs), enabling low-impedance control of each joint in the lower body via linear to rotary and parallel mechanisms. We present a distributed joint impedance control framework that leverages a custom dual-axis motor controller to track position, velocity, and torque setpoints for each pair of joints. The required actuator forces are tracked using an inner force control loop combining feedforward and PID control with a model-based disturbance observer (DOB). Unlike previous approaches, we utilize an inverse plant model based on the open-loop actuator dynamics to simplify tuning of the cascaded controller by decoupling DOB estimates from the inner loop gains. The effectiveness of the proposed approach is verified through trajectory tracking and dynamic walking experiments conducted on the THOR humanoid utilizing a complementary optimization-based whole-body controller.

Series elastic actuators are frequently modeled using a conventional lumped mass model which has remained mostly unchanged since their introduction almost two decades ago. The lumped model has served well for early development but more descriptive models are now needed for new actuator designs and control approaches. In this paper we propose a new unlumped model specifically for linear series elastic actuators which uses a rack & pinion conceptualization to intuitively depict the mechanics of a linear ball screw drive. Results from hardware experiments are presented and compared to the predicted simulation results for both the conventional model and the new unlumped model. The results demonstrate that the new unlumped model is significantly more representative of the true actuator dynamics.