Daily Arxiv

This is a page that curates AI-related papers published worldwide.
All content here is summarized using Google Gemini and operated on a non-profit basis.
Copyright for each paper belongs to the authors and their institutions; please make sure to credit the source when sharing.

VolleyBots: A Testbed for Multi-Drone Volleyball Game Combining Motion Control and Strategic Play

Created by
  • Haebom

Author

Zelai Xu, Ruize Zhang, Chao Yu, Huining Yuan, Xiangmin Yi, Shilong Ji, Chuqi Wang, Wenhao Tang, Feng Gao, Wenbo Ding, Xinlei Chen, Yu Wang

Outline

This paper presents VolleyBots, a robotic sports testbed where multiple drones cooperate and compete in a volleyball game. VolleyBots is a platform that integrates three features: competitive and cooperative gameplay, turn-based interaction structure, and agile 3D maneuvering. Drones are challenged to coordinate with their teammates and anticipate and respond to the tactics of the opposing team. Turn-based interaction requires precise timing, accurate state prediction, and management of long-term temporal dependencies, while agile 3D maneuvering requires rapid acceleration, sharp turns, and accurate 3D positioning despite the lack of quadrotor drive. In this paper, we present a comprehensive set of tasks ranging from single-drone training to multi-drone cooperation and competition tasks, and a baseline evaluation of representative multi-agent reinforcement learning (MARL) and game theory algorithms. Simulation results show that on-policy reinforcement learning (RL) methods outperform off-policy methods in single-agent tasks, but both methods struggle in complex tasks that combine motor control and strategic play. We also design a hierarchical policy that achieves a win rate of 69.5% over the strongest baseline in a 3v3 task, highlighting its potential as an effective solution for addressing complex interactions between low-level control and high-level strategy.

Takeaways, Limitations

Takeaways:
We present VolleyBots, a new testbed for evaluating intelligence implemented through robotic sports.
Demonstrates the potential of hierarchical policies as an effective solution to complex problems combining motor control and strategic play.
On-policy RL methods outperform single-agent tasks.
Limitations:
Both on-policy and off-policy RL methods struggle with complex tasks that combine motor control and strategic play.
Presenting results in a simulated environment rather than a real environment.
Further research is needed to determine whether the proposed hierarchical policy is effective in all complex situations.
👍