Agile Mobility with Rapid Online Adaptation via Meta-learning and Uncertainty-aware MPPI

TL;DR

The paper tackles rapid online adaptation for agile mobile robots operating at the limit under uncertain friction and delays. It proposes a meta-learning-based initialization to a end-to-end dynamics model, combined with an ensemble-based uncertainty estimate and an uncertainty-aware MPPI controller, enabling fast adaptation with few on-device samples. Through numeric simulation, Unity-based simulations, and hardware experiments, the approach matches or nears domain-specific controllers while maintaining strong generalization across diverse scenarios. The findings offer a practical path to robust, real-time control for wheel-based robots in changing environments, with broad applicability to other robots and tasks. The key contributions include meta-learning data generation, ensemble uncertainty, and online adaptation integrated with MPPI.

Abstract

Modern non-linear model-based controllers require an accurate physics model and model parameters to be able to control mobile robots at their limits. Also, due to surface slipping at high speeds, the friction parameters may continually change (like tire degradation in autonomous racing), and the controller may need to adapt rapidly. Many works derive a task-specific robot model with a parameter adaptation scheme that works well for the task but requires a lot of effort and tuning for each platform and task. In this work, we design a full model-learning-based controller based on meta pre-training that can very quickly adapt using few-shot dynamics data to any wheel-based robot with any model parameters, while also reasoning about model uncertainty. We demonstrate our results in small-scale numeric simulation, the large-scale Unity simulator, and on a medium-scale hardware platform with a wide range of settings. We show that our results are comparable to domain-specific well-engineered controllers, and have excellent generalization performance across all scenarios.

Figures (4)

• Figure 1: Overview of our approach. Phase 1 involves pre-training a model initialization from numeric sim data ($\sim10$M) which can quickly adapt to any vehicle platform with few-shot dynamic data. Phase 2 involves fine-tuning the model from very few dynamic data ($\sim300$) on the platform. Phase 3 involves deploying the learned model online for control using uncertainty-aware MPPI while adapting the model in real-time from online dynamic data.
• Figure 2: Lateral errors for numeric sim experiment for $10$ environments with random model parameters. Please refer to Section \ref{['sec:c5_results']} for definitions of baselines a-g.
• Figure 3: Predicted $\omega$ comparison for numeric sim experiment
• Figure 4: Unity sim environment snapshots