MIND-Stack: Modular, Interpretable, End-to-End Differentiability for Autonomous Navigation
Felix Jahncke, Johannes Betz
TL;DR
The paper tackles the challenge of achieving both interpretability and learning in autonomous navigation by introducing MIND-Stack, a modular, end-to-end differentiable stack that integrates a LiDAR-based localization module with a traditional Stanley Controller. It demonstrates end-to-end optimization where the upstream localization module is trained to minimize downstream control loss, while preserving interpretability through intermediate state representations. The approach shows strong performance advantages over state-of-the-art baselines in simulation and real-world embedded deployment, and gains from jointly training localization and controller. The work highlights sim-to-real transfer and outlines a path toward extending the differentiable framework to additional autonomous driving modules such as perception, prediction, and planning, with potential improvements for dynamic obstacle handling.
Abstract
Developing robust, efficient navigation algorithms is challenging. Rule-based methods offer interpretability and modularity but struggle with learning from large datasets, while end-to-end neural networks excel in learning but lack transparency and modularity. In this paper, we present MIND-Stack, a modular software stack consisting of a localization network and a Stanley Controller with intermediate human interpretable state representations and end-to-end differentiability. Our approach enables the upstream localization module to reduce the downstream control error, extending its role beyond state estimation. Unlike existing research on differentiable algorithms that either lack modules of the autonomous stack to span from sensor input to actuator output or real-world implementation, MIND-Stack offers both capabilities. We conduct experiments that demonstrate the ability of the localization module to reduce the downstream control loss through its end-to-end differentiability while offering better performance than state-of-the-art algorithms. We showcase sim-to-real capabilities by deploying the algorithm on a real-world embedded autonomous platform with limited computation power and demonstrate simultaneous training of both the localization and controller towards one goal. While MIND-Stack shows good results, we discuss the incorporation of additional modules from the autonomous navigation pipeline in the future, promising even greater stability and performance in the next iterations of the framework.