Synthesis of Biologically Realistic Human Motion Using Joint Torque Actuation (1904.13041v2)

Abstract: Using joint actuators to drive the skeletal movements is a common practice in character animation, but the resultant torque patterns are often unnatural or infeasible for real humans to achieve. On the other hand, physiologically-based models explicitly simulate muscles and tendons and thus produce more human-like movements and torque patterns. This paper introduces a technique to transform an optimal control problem formulated in the muscle-actuation space to an equivalent problem in the joint-actuation space, such that the solutions to both problems have the same optimal value. By solving the equivalent problem in the joint-actuation space, we can generate human-like motions comparable to those generated by musculotendon models, while retaining the benefit of simple modeling and fast computation offered by joint-actuation models. Our method transforms constant bounds on muscle activations to nonlinear, state-dependent torque limits in the joint-actuation space. In addition, the metabolic energy function on muscle activations is transformed to a nonlinear function of joint torques, joint configuration and joint velocity. Our technique can also benefit policy optimization using deep reinforcement learning approach, by providing a more anatomically realistic action space for the agent to explore during the learning process. We take the advantage of the physiologically-based simulator, OpenSim, to provide training data for learning the torque limits and the metabolic energy function. Once trained, the same torque limits and the energy function can be applied to drastically different motor tasks formulated as either trajectory optimization or policy learning. Codebase: https://github.com/jyf588/lrle and https://github.com/jyf588/lrle-rl-examples

• The paper introduces a method that transforms muscle-based control problems into joint-actuation models to achieve realistic human motion.
• It employs nonlinear, state-dependent torque limits and metabolic energy function transformations to retain biomechanical plausibility.
• The approach demonstrates significant computational efficiency in trajectory optimization and deep reinforcement learning for motion synthesis.

Synthesis of Biologically Realistic Human Motion Using Joint Torque Actuation

The paper "Synthesis of Biologically Realistic Human Motion Using Joint Torque Actuation" addresses a significant challenge in character animation and robotics: generating human-like motion using computationally efficient joint-actuation models. Conventional joint-actuation models simplify human motion by generating torques independently at each joint. However, they often lead to torque patterns that are physiologically implausible. On the other hand, muscle-actuation models that simulate dynamics through muscles and tendons yield realistic human motions but pose considerable computational complexity. This work introduces a method to bridge the gap between these two approaches by transforming muscle-based control problems into the joint-actuation space while retaining the physiological realism of motions.

The authors propose a technique that transforms an optimal control problem formulated in the muscle-actuation space to an equivalent problem in the joint-actuation space. This transformation enables the synthesis of realistic human motion with simple modeling and rapid computation. The method involves transforming constant muscle activation bounds into nonlinear, state-dependent torque limits in the joint-actuation space. Additionally, the paper presents a novel approach for transforming the metabolic energy functions defined on muscle activations into nonlinear functions of joint torques, joint configuration, and joint velocity.

Key numerical results include successful applications to trajectory optimization and policy learning using deep reinforcement learning (DRL) methods. The paper demonstrates the efficiency of its method in generating realistic jumping and swinging motions through trajectory optimization and effective walking and running policies using DRL. The models generated by the proposed method show significant improvements in computational efficiency, offering visually comparable results to those produced by muscle-actuated models while drastically reducing computation time.

Implications of this work are profound for fields like character animation, robotics, and biomechanics. Practically, it offers a framework to leverage the physiological accuracies of muscle models without incurring the computational exhaustiveness. Theoretically, it provides a foundation for new research avenues in biomechanics and physiologically accurate modeling techniques, especially useful for training autonomous agents in simulation environments. The potential application of this methodology to broader DRL scenarios could pave the way for creating robust and efficient learning algorithms that utilize anatomically realistic models of human motion.

In the future, integrating muscle co-contraction models and tendon compliance, which are not considered in this paper, could further enhance the physiological realism of the generated motions. Additionally, exploring variations in energy expenditure models could refine muscle-like behavior synthesis further.

In summary, this paper presents a technically sound method to approximate muscle-actuated models with joint-actuation models, maintaining the essential biophysical characteristics essential for many practical applications in animation and robotics.