🚀 Best Systems Paper Finalist at RSS25!!!

I’m excited to share that our latest research has been selected as a Best Systems Paper finalist at RSS 2025!

In this work, we propose a perceptive forward dynamics model that enables safe and robust navigation by predicting a robot’s future state up to 5 seconds ahead, based on its potential actions. What makes our approach stand out is its embodiment-awareness—the model goes beyond rigid-body assumptions and adapts to the specific characteristics of the platform it’s controlling.

We trained the model using a hybrid strategy that combines simulated and real-world data, allowing it to generalize across environments. When paired with an MPPI planner, it achieves strong performance in both indoor and outdoor scenarios.

📄 Paper: https://arxiv.org/pdf/2504.19322 🌐 Project: https://leggedrobotics.github.io/fdm.github.io/

Huge thanks to Jonas Frey — this collaboration was key to making the project a success.

If you’re working on forward models, planning, or safe navigation—or just curious—feel free to reach out. Always happy to chat!