Physics-based motion capture imitation with deep reinforcement learning

Chentanez, Nuttapong; Müller, Matthias; Macklin, Miles; Makoviychuk, Viktor; Jeschke, Stefan

doi:10.1145/3274247.3274506

Cited by 68 publications

(47 citation statements)

References 34 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Although physics-based methods has shown promising results for individual characters performing a wide variety of behaviors, there exist only a few studies for multi-character animations. Zordan et al [2002;2005] proposed motion capture-driven simulated characters that can react to external perturbations, where two-player interactions such as fighting, [Berseth et al 2018;Peng et al 2017;, such as imitation [Bergamin et al 2019;Chentanez et al 2018;Peng et al 2018;Won et al 2020;Won and Lee 2019], and such as other skills [Clegg et al 2018;Hodgins 2017, 2018;Xie et al 2020]. However, the number of studies on multi-characters is still limited.…”

Section: Physics-based Character Animationmentioning

confidence: 99%

Control strategies for physically simulated characters performing two-player competitive sports

2021

View full text Add to dashboard Cite

Fig. 1. Characters performing two-player competitive sports such as boxing (left) and fencing (right) using learned control strategies.In two-player competitive sports, such as boxing and fencing, athletes often demonstrate efficient and tactical movements during a competition. In this paper, we develop a learning framework that generates control policies for physically simulated athletes who have many degrees-of-freedom. Our framework uses a two step-approach, learning basic skills and learning boutlevel strategies, with deep reinforcement learning, which is inspired by the way that people how to learn competitive sports. We develop a policy model based on an encoder-decoder structure that incorporates an autoregressive latent variable, and a mixture-of-experts decoder. To show the effectiveness of our framework, we implemented two competitive sports, boxing and fencing, and demonstrate control policies learned by our framework that can generate both tactical and natural-looking behaviors. We also evaluate the control policies with comparisons to other learning configurations and with ablation studies.

show abstract

Section: Physics-based Character Animationmentioning

confidence: 99%

Control strategies for physically simulated characters performing two-player competitive sports

2021

View full text Add to dashboard Cite

show abstract

“…This strategy has been effective for imitating individual motion clips, but it can be difficult to scale to datasets containing multiple disparate motions, as it may not be possible to synchronize and align multiple reference motions according to a single-phase variable. Recent methods have extended these tracking-based techniques to larger motion datasets by explicitly providing target poses from the reference motion that is being tracked as inputs to the controller [Bergamin et al 2019;Chentanez et al 2018;Won et al 2020]. This then allows a controller to imitate different motions depending on the input target poses.…”

Section: Related Workmentioning

confidence: 99%

“…All features are recorded in the character's local coordinate system. Unlike previous systems, which synchronize the policy with a particular reference motion by including additional phase information in the state, such as scalar phase variables Peng et al 2018a] or target poses [Bergamin et al 2019;Chentanez et al 2018;Won et al 2020], our policies are not trained to explicitly imitate any specific motion from the dataset. Therefore, no such synchronization or phase information is necessary.…”

Section: States and Actionsmentioning

confidence: 99%

Amp

et al. 2021

View full text Add to dashboard Cite

system produces high-quality motions that are comparable to those achieved by state-of-the-art tracking-based techniques, while also being able to easily accommodate large datasets of unstructured motion clips. Composition of disparate skills emerges automatically from the motion prior, without requiring a high-level motion planner or other task-specific annotations of the motion clips. We demonstrate the effectiveness of our framework on a diverse cast of complex simulated characters and a challenging suite of motor control tasks. CCS Concepts: • Computing methodologies → Procedural animation; Adversarial learning; Control methods.

show abstract

“…However, although inertial sensors were used, the number of inertial motion sensors was lower than other inertial motion sensors, resulting in lower data extraction. Other studies have analyzed motion-capture data stored in databases using machine learning [13,14]; such studies have also collected motion data using optical systems, resulting in inaccurate and unreliable data. To date, most studies of human motion using databases have stored data captured with markerless optical systems.…”

Section: Human-motion Databasementioning

confidence: 99%

Development of a human-motion database for intelligent wearable robots

2021

ilkonline

View full text Add to dashboard Cite

Human-motion data need to be stored and processed to manage wearable robots. Given the considerable data volume associated with sensor data, we designed a database that efficiently stores sensor data from human motion and can be used by intelligent wearable robots. We designed a database for sensor data from human motions for use in intelligent wearable robots. We used functions to preprocess and transform the data into JSON files. Additionally, we created a function to store and process data files in the database. A database of human-motion data is necessary to determine whether robot movements (e.g., by wearable robots or robot simulations) are similar to those of humans. The most common methods for collecting motion data from people have been markerless optical systems. However, these methods are less accurate than motion capture using a non-optical system. We designed and implemented a database to collect and store accurate human-motion data using a non-optical inertial motion-capture system. Our database can be applied to walking robots used for people with paraplegic disabilities who need walking assistance, and to robots and equipment that mimic human movements for people working in high-risk environments. Our system can also be applied to humanoid robotics.

show abstract

Physics-based motion capture imitation with deep reinforcement learning

Cited by 68 publications

References 34 publications

Control strategies for physically simulated characters performing two-player competitive sports

Control strategies for physically simulated characters performing two-player competitive sports

Amp

Development of a human-motion database for intelligent wearable robots

Contact Info

Product

Resources

About