Multi-agent reinforcement learning for character control

Li, Cheng; Fussell, Levi; Komura, Taku

doi:10.1007/s00371-021-02269-1

Cited by 2 publications

(1 citation statement)

References 32 publications

(32 reference statements)

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Data-driven approaches select the optimal actions available in a database of examples, using game tree based methods [37,38]. However, due to the simplicity of the rules and the high computational complexity, the intelligence of rule-based simulated characters is too limited to handle stylized interactions [16].…”

Section: Introductionmentioning

confidence: 99%

Maaip

Younes,

Kijak,

Kulpa

et al. 2023

Proc. ACM Comput. Graph. Interact. Tech.

View full text Add to dashboard Cite

Simulating realistic interaction and motions for physics-based characters is of great interest for interactive applications, and automatic secondary character animation in the movie and video game industries. Recent works in reinforcement learning have proposed impressive results for single character simulation, especially the ones that use imitation learning based techniques. However, imitating multiple characters interactions and motions requires to also model their interactions. In this paper, we propose a novel Multi-Agent Generative Adversarial Imitation Learning based approach that generalizes the idea of motion imitation for one character to deal with both the interaction and the motions of the multiple physics-based characters. Two unstructured datasets are given as inputs: 1) a single-actor dataset containing motions of a single actor performing a set of motions linked to a specific application, and 2) an interaction dataset containing a few examples of interactions between multiple actors. Based on these datasets, our system trains control policies allowing each character to imitate the interactive skills associated with each actor, while preserving the intrinsic style. This approach has been tested on two different fighting styles, boxing and full-body martial art, to demonstrate the ability of the method to imitate different styles.

show abstract

Section: Introductionmentioning

confidence: 99%

Maaip

Younes,

Kijak,

Kulpa

et al. 2023

Proc. ACM Comput. Graph. Interact. Tech.

View full text Add to dashboard Cite

show abstract

A Hybrid Multi-Agent Reinforcement Learning Approach for Spectrum Sharing in Vehicular Networks

Jamal,

Ullah,

Naeem

et al. 2024

Future Internet

View full text Add to dashboard Cite

Efficient spectrum sharing is essential for maximizing data communication performance in Vehicular Networks (VNs). In this article, we propose a novel hybrid framework that leverages Multi-Agent Reinforcement Learning (MARL), thereby combining both centralized and decentralized learning approaches. This framework addresses scenarios where multiple vehicle-to-vehicle (V2V) links reuse the frequency spectrum preoccupied by vehicle-to-infrastructure (V2I) links. We introduce the QMIX technique with the Deep Q Networks (DQNs) algorithm to facilitate collaborative learning and efficient spectrum management. The DQN technique uses a neural network to approximate the Q value function in high-dimensional state spaces, thus mapping input states to (action, Q value) tables that facilitate self-learning across diverse scenarios. Similarly, the QMIX is a value-based technique for multi-agent environments. In the proposed model, each V2V agent having its own DQN observes the environment, receives observation, and obtains a common reward. The QMIX network receives Q values from all agents considering individual benefits and collective objectives. This mechanism leads to collective learning while V2V agents dynamically adapt to real-time conditions, thus improving VNs performance. Our research finding highlights the potential of hybrid MARL models for dynamic spectrum sharing in VNs and paves the way for advanced cooperative learning strategies in vehicular communication environments. Furthermore, we conducted an in-depth exploration of the simulation environment and performance evaluation criteria, thus concluding in a comprehensive comparative analysis with cutting-edge solutions in the field. Simulation results show that the proposed framework efficiently performs against the benchmark architecture in terms of V2V transmission probability and V2I peak data transfer.

show abstract

Multi-agent reinforcement learning for character control

Cited by 2 publications

References 32 publications

Maaip

Maaip

A Hybrid Multi-Agent Reinforcement Learning Approach for Spectrum Sharing in Vehicular Networks

Contact Info

Product

Resources

About