← ArchivePaper2019
Learning Body Shape Variation in Physics-based Characters
Parametric physics-based controller generalizes locomotion across characters with different heights, weights, and body proportions.
Abstract
Recently, deep reinforcement learning (DRL) has attracted great attention in designing controllers for physics-based characters. Despite the recent success of DRL, the learned controller is viable for a single character. Changes in body size and proportions require learning controllers from scratch. In this paper, we present a new method of learning parametric controllers for body shape variation. A single parametric controller enables us to simulate and control various characters having different heights, weights, and body proportions. The users are allowed to create new characters through body shape parameters, and they can control the characters immediately. Our characters can also change their body shapes on the fly during simulation. The key to the success of our approach includes the adaptive sampling of body shapes that tackles the challenges in learning parametric controllers, which relies on the marginal value function that measures control capabilities of body shapes. We demonstrate parametric controllers for various physically simulated characters such as bipeds, quadrupeds, and underwater animals.
How to read this
- Category
- Method: parametric DRL controller for body-shape variation
- Contributions
- Learns a single parametric physics-based controller that generalizes locomotion across heights, weights and body proportions
- Lets users author new characters via body-shape parameters and control them immediately, including changing shape on the fly during simulation
- Introduces adaptive sampling of body shapes guided by a marginal value function to make parametric control learnable
- Context
- Extends deep-reinforcement-learning character control, building directly on Peng et al.'s 'DeepMimic' (2018), generalizing from a single-character controller to a body-shape-parametric one.Builds on: DeepMimic: Example-Guided Deep Reinforcement Learning of Physics-Based Character Skills
- Correctness
- The key assumption is that one controller can span a continuous body-shape space if shapes are sampled adaptively by control capability; it is demonstrated on bipeds, quadrupeds and underwater animals, though performance still hinges on the sampling strategy and the chosen shape parameterization.
- Clarity
- Clear motivation, with depth in the DRL machinery; a first pass conveys the goal, while the adaptive sampling and marginal value function need a second pass.
- How to read it
- First pass for the parametric-controller idea and why naive training fails; second pass on the adaptive body-shape sampling and marginal value function, which are the technical heart.
Builds on
Built upon by
Nothing yet.
Related work
- DeepMimic: Example-Guided Deep Reinforcement Learning of Physics-Based Character Skills 2018 / SIGGRAPH
- Physics-Based Motion Retargeting from Sparse Inputs 2023 / SCA
- Physics-Based Character Controllers Using Conditional VAEs 2022 / SIGGRAPH
- Character Controllers Using Motion VAEs 2020 / 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 →