GDC is part of the Informa Tech Division of Informa PLC

This site is operated by a business or businesses owned by Informa PLC and all copyright resides with them. Informa PLC's registered office is 5 Howick Place, London SW1P 1WG. Registered in England and Wales. Number 8860726.

View, browse and sort the ever-growing list of sessions by pass type, track, and format. With this Session Viewer, you can view GDC 2023 session details, speakers and share your favorites via social media. You will be able to build your schedule and access it during the show via export or with the Mobile App, once live. Sessions do fill up and seating is first come, first serve, so arrive early to sessions that you would like to attend.

Multi-Agent Reinforcement Learning with 'Roller Champions'

Eva Raggini  (Machine Learning Programmer, Ubisoft)

Pass Type: All Access Pass, Core Pass

Topic: Design, Programming

Format: Lecture

Vault Recording: TBD

Audience Level: Advanced

Sports video games like Roller Champions usually require more-advanced AI teammates with real-time strategic collaboration, and more complex and realistic agent interactions. In this context, it is essential to elaborate on less predictable behaviors closer to human actions to keep the player's interest and immersion.

This talk aims to unveil the multiple challenges facing the introduction of deep reinforcement learning (Deep RL) in a real-world game production that must deliver high-end results in a multi-player environment where agents must effectively collaborate with real players.

Eva Raggini, Machine Learning Programmer at Ubisoft, will show how the team achieved the modeling of multiple complex behaviors using multi-agent reinforcement learning (MARL), how they overcame a range of difficulties within complex training environments, their approach to handling collaboration between agents and players, and why they incorporated changes in the production workflow to rely more heavily on machine learning inference.

Takeaway

Attendees will gain useful insights into managing specific reinforcement learning issues in a multi-agent system environment. The goal is to help future machine learning programmers identify the challenges related to this technology, enlightening the workflow steps to model complex NPC behaviors required to achieve a production-ready quality.

Intended Audience

This is for ML/AI programmers interested in reinforcement learning usage and technological advancement in game production.