{"667599":{"#nid":"667599","#data":{"type":"news","title":"Like Humans and Animals, AI Agents Find Their Way Through Memory","body":[{"value":"\u003Cp\u003EMemory may be just as important to artificial intelligence (AI) agents in creating \u2018mental maps\u2019 as it is to humans and animals.\u003C\/p\u003E\r\n\r\n\u003Cp\u003EA recent paper authored by Georgia Tech researchers makes a surprising discovery \u2014 blind AI agents use memory to create maps and navigate through their surrounding environment.\u003C\/p\u003E\r\n\r\n\u003Cp\u003EErik Wijmans, the lead author of the paper, said the idea for his research began by asking if AI agents might mimic human and animal behavior in how they navigate and adjust to their environments.\u003C\/p\u003E\r\n\r\n\u003Cp\u003E\u201cHumans and animals navigate with some type of spatial representation \u2014 what is commonly referred to as a cognitive map,\u201d Wijmans said. \u201cSo, we were wondering how AI agents navigate and if it\u2019s similar to that.\u003C\/p\u003E\r\n\r\n\u003Cp\u003E\u201cThe first question we asked was, \u2018Is memory important to these agents?\u0027 It is. They tend to remember at least the past thousand interactions with their environment.\u201d\u003C\/p\u003E\r\n\r\n\u003Cp\u003EWijmans completed his Ph.D. in computer science in 2022 and is currently a research scientist at Apple.\u003C\/p\u003E\r\n\r\n\u003Cp\u003EWijmans created blind AI agents and trained them by dropping them into the floorplans of more than 500 houses with the goal of navigating from one area of the house to another area. The only sense it had to work with was egomotion \u2014 the ability to know how far it has moved.\u003C\/p\u003E\r\n\r\n\u003Cp\u003EThe agent bumped its way around from room to room, backtracking as needed, before finding its destination. Wijmans then created a second probe agent that was injected with the memories of the first agent. The probe agent used the memory of the original agent to take shortcuts to quickly reach its objective.\u003C\/p\u003E\r\n\r\n\u003Cp\u003E\u201cIt\u2019s surprising that they can do this without vision because they\u2019re in an unknown environment that they\u2019ve never seen before, so they have to figure out how to navigate in that environment and also figure out the structure of it,\u201d Wijmans said.\u003C\/p\u003E\r\n\r\n\u003Cp\u003E\u201cThis is a result that shows that our hypothesis is true, or at the very least along the right direction. We took an agent and put it in a complex environment and trained it for a task that requires it to interact with that environment, and the result was mapping.\u201d\u003C\/p\u003E\r\n\r\n\u003Cp\u003EWijman\u2019s paper,\u0026nbsp;\u003Cem\u003EEmergence of Maps in the Memories of Blind Navigation Agents\u003C\/em\u003E, is one of four outstanding paper award winners for the 2023 International Conference on Learning Representations, which is being held May 1-5 in Kigali, Rwanda. His research was also recognized by the Georgia Tech chapter of Sigma Xi (The Scientific Research Society) and received a 2023 GT Sigma XI Best Ph.D. Thesis Award.\u003C\/p\u003E\r\n\r\n\u003Cp\u003EWijmans is advised by School of Interactive Computing Distinguished Professor Irfan Essa and Associate Professor Dhruv Batra.\u003C\/p\u003E\r\n\r\n\u003Cp\u003E\u201cErik makes fundamental contributions to multiple sub-areas of AI, including reinforcement learning, robotics, and embodied perception,\u201d Batra said. \u201cHis hypothesis is a bold one \u2014 that intelligence emerges via large-scale learning by an embodied agent accomplishing goals in a rich 3D environment.\u201d\u003C\/p\u003E\r\n\r\n\u003Cp\u003EIn his paper, Wijmans describes mapping as an emerging phenomenon. Neural network models for navigation have performed well despite not containing any explicit mapping modules.\u003C\/p\u003E\r\n\r\n\u003Cp\u003EWijman\u2019s AI agents showed a 95% success rate when they used memory to navigate, whereas memoryless agents failed entirely. This seems to suggest that agents create mental maps as a natural part of learning to navigate.\u003C\/p\u003E\r\n\r\n\u003Cp\u003E\u201cThe results were so initially surprising, that my first gut instinct was that we had done something wrong in our experimental design,\u201d he said.\u003C\/p\u003E\r\n\r\n\u003Cp\u003E\u201cThis is a work with a very complex body of experiments that tie together a single narrative,\u201d he said. \u201cThis is a challenging thing to do. When you\u2019re trying to test whether something involves memory, you must come up with ideas of what to test for and how to test for that. You must make each experiment as precise as possible to not get false positives, and that involves considerable experimental design and effort.\u201d\u003C\/p\u003E\r\n\r\n\u003Cp\u003EWijmans said he made it as difficult as possible for the agent to reach its goal, removing vision, audio, olfactory, haptic, and magnetic sensing and gave it no bias toward mapping. It had no supervision or any kind of outside help.\u003C\/p\u003E\r\n\r\n\u003Cp\u003E\u201cSurprisingly, even under these deliberately harsh conditions, we find the emergence of map-like spatial representations in the agent\u2019s non-spatial unstructured memory. It not only successfully navigates to the goal but also exhibits intelligent behavior like taking shortcuts, following walls, and detecting collisions.\u201d\u003C\/p\u003E\r\n\r\n\u003Cp\u003EThe discovery also suggests that AI, humans, and animals all share a natural characteristic of problem solving and navigation.\u003C\/p\u003E\r\n\r\n\u003Cp\u003E\u201cThe one link that we can make is the idea of convergent evolution, which is where you see the same mechanism evolve multiple times in species that have no common ancestor that shares that mechanism,\u201d Wijmans said. \u201cMammals build maps, insects build maps, and now AI agents build maps. So perhaps mapping is the natural solution to navigation.\u201d\u003C\/p\u003E\r\n","summary":"","format":"limited_html"}],"field_subtitle":"","field_summary":[{"value":"\u003Cp\u003EA recent paper authored by Georgia Tech researchers makes a surprising discovery \u2014 blind AI agents use memory to create maps and navigate through their surrounding environment.\u003C\/p\u003E\r\n","format":"limited_html"}],"field_summary_sentence":[{"value":"A recent paper authored by Georgia Tech researchers makes a surprising discovery \u2014 blind AI agents use memory to create maps and navigate through their surrounding environment."}],"uid":"32045","created_gmt":"2023-05-02 13:23:26","changed_gmt":"2023-05-02 13:28:07","author":"Ben Snedeker","boilerplate_text":"","field_publication":"","field_article_url":"","dateline":{"date":"2023-05-02T00:00:00-04:00","iso_date":"2023-05-02T00:00:00-04:00","tz":"America\/New_York"},"extras":[],"hg_media":{"670706":{"id":"670706","type":"image","title":"Erik Wijmans, Irfan Essa","body":null,"created":"1683033816","gmt_created":"2023-05-02 13:23:36","changed":"1683033816","gmt_changed":"2023-05-02 13:23:36","alt":"Georgia Tech Ph.D. student Erik Wijmans and Distinguished Professor Irfan Essa","file":{"fid":"253618","name":"Erik Wijmans, Irfan Essa_86A9563.jpeg","image_path":"\/sites\/default\/files\/2023\/05\/02\/Erik%20Wijmans%2C%20Irfan%20Essa_86A9563.jpeg","image_full_path":"http:\/\/hg.gatech.edu\/\/sites\/default\/files\/2023\/05\/02\/Erik%20Wijmans%2C%20Irfan%20Essa_86A9563.jpeg","mime":"image\/jpeg","size":38004,"path_740":"http:\/\/hg.gatech.edu\/sites\/default\/files\/styles\/740xx_scale\/public\/2023\/05\/02\/Erik%20Wijmans%2C%20Irfan%20Essa_86A9563.jpeg?itok=GUWuL3qy"}}},"media_ids":["670706"],"groups":[{"id":"576481","name":"ML@GT"},{"id":"50876","name":"School of Interactive Computing"}],"categories":[{"id":"8862","name":"Student Research"},{"id":"153","name":"Computer Science\/Information Technology and Security"},{"id":"152","name":"Robotics"},{"id":"135","name":"Research"}],"keywords":[{"id":"187812","name":"artificial intelligence (AI)"}],"core_research_areas":[{"id":"39501","name":"People and Technology"},{"id":"39521","name":"Robotics"}],"news_room_topics":[],"event_categories":[],"invited_audience":[],"affiliations":[],"classification":[],"areas_of_expertise":[],"news_and_recent_appearances":[],"phone":[],"contact":[{"value":"\u003Cp\u003ENathan Deen\u003Cbr \/\u003E\r\nCommunications Officer I\u003Cbr \/\u003E\r\nSchool of Interactive Computing\u003Cbr \/\u003E\r\nnathan.deen@cc.gatech.edu\u003C\/p\u003E\r\n","format":"limited_html"}],"email":["nathan.deen@cc.gatech.edu"],"slides":[],"orientation":[],"userdata":""}}}