← ArchivePaper2019
DReCon: Data-Driven Responsive Control of Physics-Based Characters
Physics-based character controller combining motion matching for reference generation with tracking control for responsive, physically plausible animation.
Abstract
Interactive control of self-balancing, physically simulated humanoids is a long standing problem in the field of real-time character animation. While physical simulation guarantees realistic interactions in the virtual world, simulated characters can appear unnatural if they perform unusual movements in order to maintain balance. Therefore, obtaining a high level of responsiveness to user control, runtime performance, and diversity has often been overlooked in exchange for motion quality. Recent work in the field of deep reinforcement learning has shown that training physically simulated characters to follow motion capture clips can yield high quality tracking results. We propose a two-step approach for building responsive simulated character controllers from unstructured motion capture data. First, meaningful features from the data such as movement direction, heading direction, speed, and locomotion style, are interactively specified and drive a kinematic character controller implemented using motion matching. Second, reinforcement learning is used to train a simulated character controller that is general enough to track the entire distribution of motion that can be generated by the kinematic controller. Our design emphasizes responsiveness to user input, visual quality, and low runtime cost for application in video-games.
How to read this
- Category
- Method: data-driven physics-based character control
- Contributions
- Proposes a two-step controller that builds responsive simulated characters from unstructured motion capture data
- Uses motion matching as an interactively specified kinematic controller (movement direction, heading, speed, locomotion style) to generate reference motion
- Trains a reinforcement-learning tracking controller general enough to follow the full distribution of kinematically generated motion
- Context
- Combines DeepMimic-style example-guided RL tracking (Peng et al., 2018) with Motion Matching (Clavet, 2016) to balance responsiveness against physical plausibility.Builds on: DeepMimic: Example-Guided Deep Reinforcement Learning of Physics-Based Character Skills · Motion Matching and The Road to Next-Gen Animation
- Correctness
- Assumes a kinematic motion-matching layer can produce a trackable reference distribution that the RL policy then follows under simulation; results target self-balancing humanoids and real-time control, so behavior outside the captured motion distribution and tuning effort remain considerations.
- Clarity
- Clearly structured around the two stages; a first pass conveys the pipeline, with a second pass for the RL tracking and reward details.
- How to read it
- Read for the two-stage design: understand how motion matching feeds the RL tracker on a first pass, then do a second pass on the policy training if you intend to reproduce or extend the controller.
Builds on
Related work
- SuperTrack: Motion Tracking for Physically Simulated Characters Using Supervisory Signals 2021 / SIGGRAPH Asia
- Character Controllers Using Motion VAEs 2020 / SIGGRAPH
- AMP: Adversarial Motion Priors for Stylized Physics-Based Character Control 2021 / SIGGRAPH
- Near-Optimal Character Animation with Continuous Control 2007 / SIGGRAPH
Keywords
This page summarises the entry and links to its original source. The archive never hosts or redistributes the publication itself.Show it in the full archive list →