Motional Releases nuReasoning Dataset for Autonomous Driving

Motional announced in a press release the release of nuReasoning, an open dataset for training autonomous driving systems to reason through rare and complex situations. It contains 20,000 scenarios and more than 105 hours of driving video collected in Las Vegas, Pittsburgh, Los Angeles, Boston, and Singapore.

The dataset covers unusual pedestrian activity, work zones, nighttime construction, animal crossings, and limited visibility. Each event contains at least 20 seconds of video, multimodal sensor data, and human verified annotations that explain driving decisions and why alternative actions were considered unsafe.

NuReasoning includes 247,000 annotations covering spatial, decision, and counterfactual reasoning. Its Omnitag search engine lets users find data by scenario, difficulty, location, or natural language descriptions.

Motional and the UCLA Mobility Lab are also hosting a nuReasoning Challenge at the European Conference on Computer Vision in Sweden. The contest uses 1,000 private test scenarios to evaluate trajectory planning, motion planning, visual question answering, and scene reasoning, with winners due to be announced at NeurIPS in December.