[Research Internship] Improving Non-Player Character Decision Making with ML (FMNB)

Recent advances in deep learning and reinforcement learning offer promising new ways to design powerful decision making systems. While multiple works studied how to leverage large player databases to learn such systems (e.g. bots from player traces) this project proposes to focus on a data-scarce scenario: how to enrich NPC (non-player character) behaviors in video games for which little to no player or human data is available. 

In this internship you will explore and develop deep learning methods to improve NPC decision-making under data-scarce conditions. Possible application domains include fight behavior (combat AI) navigation in dynamic environments or adaptive strategy in non-combat interactions. A core challenge is bridging the gap between classical game AI techniques and modern deep learning models to produce realistic responsive and computationally feasible NPCs. 

You will also work closely with Ubisofts production and AI teams to ensure your research is grounded in real-world game constraints. 

The internship will be hosted at Ubisofts La Forge giving you access to research infrastructure: 3D game environments built for experimentation and GPU clusters. You will work alongside Ubisoft researchers and developers to iterate on prototypes with regular interactions to align research with production constraints and game design needs. 

OBJECTIVES

• Survey state-of-the-art deep learning and reinforcement learning methods for decision-making with emphasis on approaches relying on self-collected data (reinforcement learning synthetic data simulation). 
• Explore hybrid architectures that combine classical game AI (behavior trees finite-state machines path planners) with neural networks e.g. learning residuals or decision modulations. 
• Collaborate with production teams to define a concrete gameplay scenario (combat navigation cooperative AI etc.) as target for experimentation. 
• Implement prototypes in realistic 3D environments evaluate them in terms of realism responsiveness generalization and computational cost. 

Qualifications :

• You are in the final year of an engineering degree or pursuing a research master in computer science or a related field. 
• Solid foundation in algorithms probability linear algebra optimization and machine learning. 
• Experience in Python and at least one deep learning framework (PyTorch TensorFlow etc.). 
• Familiarity (or strong interest) in reinforcement learning control decision-making models and/or imitation learning. 
• Good English communication skills capable of working in an international multidisciplinary environment. 
• Passion for game AI game mechanics and interactive systems. 

Additional Information :

• Contract : 6 months Internship
• Location : role based in Bordeaux France
• Remote: We embrace a hybrid work model helping you stay connected with your team and aligned with business priorities while giving you the opportunity to maintain your work-life balance. Note that some roles are fully office-based and are not eligible for hybrid work.

Relevant Literature 

• (example of a Ubisoft initiative) 
• Efficient Active Imitation Learning with Random Network Distillation (ICLR paper done with a former intern !) 
• Progressive growing of self-organized hierarchical representations for exploration 
• Adversarial Reinforcement Learning for Procedural Content Generation 

Check out this guide to help you with your application and learn about our actions to encourage more diversity and inclusion.

Remote Work :

No

Employment Type :

Intern