AI discovers NHL 26 goalie exploits using novel RL approach

Researchers have developed a novel approach called Reward-Adaptive Iterative Discovery (RAID) to automate the testing of AI agents in video games. This method uses an iterative Reinforcement Learning (RL) approach to train multiple goal-scoring agents, aiming to discover diverse exploit strategies more efficiently than traditional RL methods. In a case study involving a development version of EA SPORTS NHL 26, RAID successfully identified six distinct scoring exploit strategies within a single experiment, mirroring the types of exploits found by human playtesters over much longer periods. AI

IMPACT Automates game testing, potentially reducing development costs and improving AI robustness.

RANK_REASON Academic paper detailing a new AI methodology. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.AI →

• EA SPORTS NHL 26
• Reinforcement Learning
• Reward-Adaptive Iterative Discovery

• paper
• product

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →