{"version":1,"type":"story","url":"https://digestai.news/story/compact-multimodal-imitation-policy-drives-hours-collision-free-in-carla","json":"https://digestai.news/story/compact-multimodal-imitation-policy-drives-hours-collision-free-in-carla.json","markdown":"https://digestai.news/story/compact-multimodal-imitation-policy-drives-hours-collision-free-in-carla.md","slug":"compact-multimodal-imitation-policy-drives-hours-collision-free-in-carla","headline":"Compact Multimodal Imitation Policy Drives Hours Collision-Free in CARLA","summary":"The paper presents a compact multimodal driving policy trained via behavioral cloning on offline expert demonstrations and evaluated in the CARLA simulator. The policy ingests five‑frame histories of RGB images, LiDAR point clouds, vehicle telemetry, and lane waypoints, outputting throttle, brake and steering commands at 20 Hz.\n\nTraining used 236,882 short windows—about 3.3 hours of driving collected from 448 separate captures—resulting in a 1.36 million‑parameter network. In closed‑loop tests the model navigated both training routes and previously unseen held‑out routes for multiple hours without collisions, and it transferred qualitatively to a different CARLA town with distinct road geometry. The authors also note occasional recovery from large trajectory deviations, though systematic recovery was not formally evaluated.\n\nAll code, the trained checkpoint, an ONNX export, sample data, and an evidence audit are released publicly, enabling reproducibility and further research on offline imitation learning for autonomous driving.","keyPoints":["1.36 M‑parameter multimodal policy trained on 3.3 h of expert data (236,882 windows)","Achieves hours of collision‑free autonomous driving on both seen and unseen CARLA routes","Public release includes code, checkpoint, ONNX model, data sample, and audit documentation"],"whyItMatters":"Demonstrates that modest‑size offline‑trained models can attain reliable closed‑loop driving, lowering barriers for research and accelerating safe autonomous‑vehicle development.","category":{"slug":"research","name":"Research","url":"https://digestai.news/category/research"},"entities":{"companies":[],"models":[],"people":[]},"firstPublishedAt":"2026-09-17T04:00:00Z","updatedAt":"2026-09-17T04:00:00Z","sourceCount":1,"hasPrimarySource":true,"sources":[{"outlet":"arXiv cs.AI","title":"Imitation Learning for Autonomous Driving in CARLA","url":"https://arxiv.org/abs/2609.17757","publishedAt":"2026-09-17T04:00:00Z","type":"primary","primary":true,"lead":true}],"sourceNotes":null,"discussions":[],"thread":null,"cite":{"text":"Digest AI, \"Compact Multimodal Imitation Policy Drives Hours Collision-Free in CARLA\", 17 September 2026, https://digestai.news/story/compact-multimodal-imitation-policy-drives-hours-collision-free-in-carla","publisher":"Digest AI","title":"Compact Multimodal Imitation Policy Drives Hours Collision-Free in CARLA","datePublished":"2026-09-17T04:00:00Z","url":"https://digestai.news/story/compact-multimodal-imitation-policy-drives-hours-collision-free-in-carla"},"generatedBy":"Written by Digest AI's editorial model from the linked sources; the sources are the record.","license":"Headlines, digests and key points are written by Digest AI and may be quoted with a link to the story page. Linked articles belong to their publishers. Terms: https://digestai.news/terms#reuse"}