What this is
National mobility networks move millions of people and goods every day under stochastic demand, aging infrastructure and tightening carbon budgets. ORAILIX builds AI methods that work with, not instead of, the operations-research playbook mobility operators already use, so improvements are deployable, interpretable, and compatible with safety-critical schedules. The work is anchored by the Chair Optimisation and AI for Mobility at École Polytechnique, co-funded by SNCF Group.
Why it matters
Pure deep learning on scheduling and routing rarely clears the bar operators demand: no formal guarantees, no explainable deviations from a plan, no robustness under distributional shift. Pure OR, conversely, scales poorly to high-dimensional sensor streams and non-stationary demand. ORAILIX works on the interface. Learning the irreducibly data-driven parts while keeping the decision layer under constraint-programming or integer-programming control. The result is methods SNCF, Safran and similar operators can ship.
Representative work
- Simulation-Driven Railway Delay Prediction, An Imitation Learning Approach (AAAI-26). Imitation learning from a dispatcher simulator predicts delay propagation on SNCF-grade data; complements the industrial scheduling stack.
- Uncertainty Quantification as a Complementary Latent Health Indicator (MIMAR 2025). Predictive maintenance for turbofan engines.
- I-GLIDE, Input Groups for Latent Health Indicators in Degradation Estimation (ECML PKDD 2025). Captures interaction effects classical degradation models miss.
- TurboSens, Reward or the Observation Stream? (RLxF workshop at ICML 2026). World-model auditing for RL policies, using a turbofan substrate where ground truth is available.
- Open RL turbofan degradation challenge. Released with the École Polytechnique course Reinforcement Learning and Autonomous Agents.
Partners
Industrial chair with SNCF Group. Participation in the IRT SystemX JNI3 programme with Safran Tech on turbofan health and digital twins. Visibility at VivaTech, the Journées Industrielles and international academic venues (AAAI, ICML, ECML PKDD).
Related news
Follow related dispatches in the news stream, especially the tech and publications channels.
← Back to ORAILIX