{"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/learning-to-fly-by-crashing","title":"Learning to Fly by Crashing","arxiv_id":"1704.05588","date":"2017-04-19","proceeding":null,"authors":["Dhiraj Gandhi","Lerrel Pinto","Abhinav Gupta"],"abstract":"How do you learn to navigate an Unmanned Aerial Vehicle (UAV) and avoid\nobstacles? One approach is to use a small dataset collected by human experts:\nhowever, high capacity learning algorithms tend to overfit when trained with\nlittle data. An alternative is to use simulation. But the gap between\nsimulation and real world remains large especially for perception problems. The\nreason most research avoids using large-scale real data is the fear of crashes!\nIn this paper, we propose to bite the bullet and collect a dataset of crashes\nitself! We build a drone whose sole purpose is to crash into objects: it\nsamples naive trajectories and crashes into random objects. We crash our drone\n11,500 times to create one of the biggest UAV crash dataset. This dataset\ncaptures the different ways in which a UAV can crash. We use all this negative\nflying data in conjunction with positive data sampled from the same\ntrajectories to learn a simple yet powerful policy for UAV navigation. We show\nthat this simple self-supervised model is quite effective in navigating the UAV\neven in extremely cluttered environments with dynamic obstacles including\nhumans. For supplementary video see: https://youtu.be/u151hJaGKUo","url_abs":"http://arxiv.org/abs/1704.05588v2","url_pdf":"http://arxiv.org/pdf/1704.05588v2.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":"learning-to-fly-by-crashing","repo_url":"https://github.com/DanielDworakowski/flot","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"tf","reach":{"status":"unanswered"}}],"tasks":[{"task_slug":"navigate","task_name":"Navigate"}],"methods":[],"datasets_introduced":[],"methods_introduced":[],"results":[],"syntology":{"atlas_url":"https://app.syntology.ai/?focus=1704.05588","mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}