New ExPhy Benchmark Improves Multi-Object Trajectory Forecasting with Physics.

Rui Wang, Yeteng Wu, Xianlin Zhang, Mengshi Qi· August 21, 2026 View original

Key takeaways

  • ExPhy is a new benchmark for evaluating AI models' ability to learn explicit physical properties in trajectory forecasting.
  • Physics-guided models like PhyODE can significantly improve prediction accuracy, especially in novel scenarios.
  • Accurate trajectory forecasting does not automatically imply accurate physical property recovery.
  • Explicitly modeling physical properties enhances model interpretability and generalization.

Who benefits

Autonomous VehiclesRoboticsLogisticsManufacturing

Summary

Researchers introduce ExPhy, a new benchmark for multi-object trajectory forecasting that explicitly labels physical properties like mass and friction. A physics-guided model, PhyODE, significantly improves forecasting accuracy, especially in out-of-distribution scenarios.

A new benchmark called ExPhy has been developed to advance multi-object trajectory forecasting by focusing on the explicit learning of physical properties. Unlike previous benchmarks, ExPhy provides 24,000 simulated physical scenes with detailed object-level labels for mass, friction, and restitution, alongside observed and future trajectories. It includes both in-distribution and out-of-distribution splits to rigorously test models' ability to generalize. To demonstrate the benchmark's utility, the researchers also introduced PhyODE, a physics-guided model designed to estimate physical properties from observed trajectories and use them for future predictions. PhyODE showed substantial improvements in long-horizon out-of-distribution forecasting, reducing error rates by over 30% compared to strong baselines. The study also highlighted that accurate trajectory prediction doesn't always mean a model has accurately recovered the underlying physical properties.

Why it matters

This research is crucial for developing more robust and interpretable AI systems in robotics and autonomous vehicles, as understanding underlying physics improves prediction accuracy and generalization to novel situations.

How to implement this in your domain

  1. 1Integrate physics-informed models into existing trajectory prediction pipelines for autonomous systems.
  2. 2Leverage the ExPhy benchmark to evaluate and compare the physical reasoning capabilities of new AI models.
  3. 3Develop new model architectures that explicitly learn and utilize physical properties for enhanced prediction and interpretability.
  4. 4Conduct stress tests on current AI models using the OOD splits of ExPhy to identify generalization weaknesses.

Original post by Rui Wang, Yeteng Wu, Xianlin Zhang, Mengshi Qi

"arXiv:2608.20009v1 Announce Type: new Abstract: Understanding object dynamics requires not only predicting future trajectories but also examining whether a model captures the physical properties that govern motion. However, existing benchmarks rarely expose object-level physical…"

View on X

Originally posted by Rui Wang, Yeteng Wu, Xianlin Zhang, Mengshi Qi on X · view source

Want to go deeper?

Turn these trends into skills with Learnijoy's hands-on AI & tech courses.

Explore courses

More in AI Research

AI ResearchAI Engineering & DevTools

Decoding Silent Reading from Non-Invasive EEG

This research demonstrates that open-vocabulary word-level and semantic information can be reliably decoded from non-invasive EEG during silent reading. Using a contrastive decoder and a large dataset from a single participant, the study shows decoding scales log-linearly with training data and extends to rare words.

Ingo Marquardt, Anthilia Alchanat, Priyanka JainAug 21, 2026
AI ResearchAI Engineering & DevTools

Exact Learning Coefficients for Singular Models

This paper presents the first deterministic algorithm for exactly computing local learning coefficients (Real Log Canonical Thresholds) for two-dimensional singular models. This breakthrough provides ground truth for calibrating sampling-based estimators and reveals algebraic structure in learning coefficients, outperforming sampling in shallow regimes.

Gr\'egoire Sergeant-Perthuis (CQSB, Sorbonne Universit\'e), Elias Tsigaridas (Ouragan Team, INRIA), Jules Tsukahara (Ouragan Team, INRIA)Aug 21, 2026
AI Engineering & DevToolsAI Research

Standardized ML Evaluation for Power System Protection

This paper proposes a standardized framework for evaluating machine learning applications in power system protection, addressing inconsistencies in current research. It defines seven critical study dimensions and instantiates the framework with a case study on fault classification and localization using a public benchmark.

Julian Oelhaf, Georg Kordowich, Paula Andrea P\'erez-Toro, Christian Bergler, Johann J\"ager, Andreas Maier, Siming BayerAug 21, 2026