{"about":{"site":"https://codewithpapers.app","non_affiliation":"Code with Papers and Syntology are not affiliated with, endorsed by, or sponsored by Papers with Code, Meta, or the pwc-archive mirror.","licence":"CC BY-SA 4.0","licence_url":"https://creativecommons.org/licenses/by-sa/4.0/legalcode","attribution":"https://codewithpapers.app/attribution","modified":"archive material modified by Syntology; see the attribution page"},"url":"/paper/inverse-reinforcement-learning-for-video","title":"Inverse reinforcement learning for video games","arxiv_id":"1810.10593","date":"2018-10-24","proceeding":null,"authors":["Aaron Tucker","Adam Gleave","Stuart Russell"],"abstract":"Deep reinforcement learning achieves superhuman performance in a range of\nvideo game environments, but requires that a designer manually specify a reward\nfunction. It is often easier to provide demonstrations of a target behavior\nthan to design a reward function describing that behavior. Inverse\nreinforcement learning (IRL) algorithms can infer a reward from demonstrations\nin low-dimensional continuous control environments, but there has been little\nwork on applying IRL to high-dimensional video games. In our CNN-AIRL baseline,\nwe modify the state-of-the-art adversarial IRL (AIRL) algorithm to use CNNs for\nthe generator and discriminator. To stabilize training, we normalize the reward\nand increase the size of the discriminator training dataset. We additionally\nlearn a low-dimensional state representation using a novel autoencoder\narchitecture tuned for video game environments. This embedding is used as input\nto the reward network, improving the sample efficiency of expert\ndemonstrations. Our method achieves high-level performance on the simple\nCatcher video game, substantially outperforming the CNN-AIRL baseline. We also\nscore points on the Enduro Atari racing game, but do not match expert\nperformance, highlighting the need for further work.","url_abs":"http://arxiv.org/abs/1810.10593v1","url_pdf":"http://arxiv.org/pdf/1810.10593v1.pdf","source":{"archive":"pwc-archive (Hugging Face), CC BY-SA 4.0","snapshot":"2025-07-28","licence_url":"https://creativecommons.org/licenses/by-sa/4.0/legalcode","row_kind":"abstracts"},"code_links":[{"paper_slug":"inverse-reinforcement-learning-for-video","repo_url":"https://github.com/HumanCompatibleAI/atari-irl","is_official":1,"mentioned_in_paper":1,"mentioned_in_github":0,"framework":"tf","reach":{"status":"ok"}}],"tasks":[{"task_slug":"continuous-control","task_name":"Continuous Control"},{"task_slug":"deep-reinforcement-learning","task_name":"Deep Reinforcement Learning"},{"task_slug":"reinforcement-learning","task_name":"Reinforcement Learning"},{"task_slug":"reinforcement-learning-1","task_name":"Reinforcement Learning (RL)"},{"task_slug":"continuous-control","task_name":"continuous-control"},{"task_slug":"reinforcement-learning-2","task_name":"reinforcement-learning"}],"methods":[],"datasets_introduced":[],"methods_introduced":[],"results":[],"syntology":{"syntology_url":"https://syntology.ai/paper/1810.10593","atlas_url":"https://app.syntology.ai/?focus=1810.10593","mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}