# Lucy-SKG: Learning to Play Rocket League Efficiently Using Deep Reinforcement Learning

**Authors:** Vasileios Moschopoulos, Pantelis Kyriakidis, Aristotelis Lazaridis, Ioannis Vlahavas  
**arXiv:** [2305.15801](https://arxiv.org/abs/2305.15801)  
**Date:** 2023-05-25  
**Summary page:** https://www.summarizepaper.com/en/paper/2305.15801v1/

## Key points

- Treating games as problems has been successful in advancing AI and led to breakthroughs
- Rocket League is a complex 3D multiplayer video game that poses significant challenges for developing efficient game-playing agents
- Lucy-SKG is a Reinforcement Learning-based model that learned how to play Rocket League in a sample-efficient manner and outperformed the two highest-ranking bots in the game
- The authors developed a reward analysis and visualization library, novel parameterizable reward shape functions, and auxiliary neural architectures for training on reward prediction and state representation tasks for enhanced efficiency in learning speed and performance
- Thorough ablation studies showed the independent effectiveness of each component of Lucy-SKG on overall performance
- Sample-efficient Reinforcement Learning techniques can be used to control complex dynamical systems under competitive team-based multiplayer conditions
- This research contributes to advancing AI techniques through its application to real-world problems such as game-playing agents, highlighting potential applications beyond entertainment purposes.

## Summary

In the field of artificial intelligence, treating games as problems has been a successful tactic for advancing AI and has led to various breakthroughs. This paper applies this strategy to study Rocket League, a widely popular 3D multiplayer video game with complex dynamics that pose significant challenges in developing efficient and high-performance game-playing agents. The authors present Lucy-SKG, a Reinforcement Learning-based model that learned how to play Rocket League in a sample-efficient manner and outperformed the two highest-ranking bots in this game. The authors' contributions include the development of a reward analysis and visualization library; novel parameterizable reward shape functions that capture the utility of complex reward types via their proposed Kinesthetic Reward Combination (KRC) technique; and design of auxiliary neural architectures for training on reward prediction and state representation tasks in an on-policy fashion for enhanced efficiency in learning speed and performance. By performing thorough ablation studies for each component of Lucy-SKG, they showed their independent effectiveness in overall performance. Furthermore, the authors demonstrate the prospects and challenges of using sample-efficient Reinforcement Learning techniques for controlling complex dynamical systems under competitive team-based multiplayer conditions. This research contributes to advancing AI techniques through its application to real-world problems such as game-playing agents. It also highlights the potential applications of AI in gaming environments beyond entertainment purposes. Overall, this study provides insights into how AI can be used effectively to solve challenging problems related to gaming environments.