# Imperative Learning: A Self-supervised Neural-Symbolic Learning Framework for Robot Autonomy

**Authors:** Chen Wang, Kaiyi Ji, Junyi Geng, Zhongqiang Ren, Taimeng Fu, Fan Yang, Yifan Guo, Haonan He, Xiangyu Chen, Zitong Zhan, Qiwei Du, Shaoshu Su, Bowen Li, Yuheng Qiu, Yi Du, Qihang Li, Yifan Yang, Xiao Lin, Zhipeng Zhao  
**arXiv:** [2406.16087](https://arxiv.org/abs/2406.16087)  
**Date:** 2024-06-23  
**Summary page:** https://www.summarizepaper.com/en/arxiv-id/2406.16087v1/

## Key points

- Data-driven methods like reinforcement and imitation learning have shown success in robot autonomy but face challenges with large datasets and adaptability to dynamic environments.
- A new self-supervised neural-symbolic computational framework called imperative learning (IL) has been introduced to address these limitations.
- IL leverages symbolic reasoning to enhance generalization abilities for robots and comprises three key components: a neural module, a reasoning engine, and a memory system.
- Formulating IL as a special bilevel optimization enables reciprocal learning among modules, reducing the need for extensive labeled data and leveraging symbolic reasoning related to logical reasoning, physical principles, geometric analysis, etc.
- Various optimization techniques for IL have been discussed across five distinct robot autonomy tasks showing effectiveness in path planning, rule induction, optimal control, visual odometry, and multi-robot routing.
- Experiments with IL have demonstrated significant improvements in robot autonomy capabilities with potential impacts extending beyond robotics into diverse domains.

## Summary

, , , , 

In the field of robot autonomy, data-driven methods like reinforcement and imitation learning have shown significant success. However, their reliance on large datasets poses challenges in adapting to dynamic environments and can be costly to collect. To address these limitations, a new self-supervised neural-symbolic (NeSy) computational framework called imperative learning (IL) has been introduced. IL leverages symbolic reasoning to enhance generalization abilities for robots. The IL framework comprises three key components: a neural module, a reasoning engine, and a memory system. By formulating IL as a special bilevel optimization (BLO), reciprocal learning is enabled among these modules. This approach mitigates the need for extensive labeled data common in data-driven methods and capitalizes on symbolic reasoning related to logical reasoning, physical principles, geometric analysis, and more. Various optimization techniques have been discussed for IL, demonstrating their effectiveness across five distinct robot autonomy tasks including path planning, rule induction, optimal control, visual odometry, and multi-robot routing. Through experiments conducted with IL, it has been shown to significantly improve robot autonomy capabilities. The potential impact of this framework extends beyond robotics into diverse domains, paving the way for further research advancements in autonomous systems. Authors involved in this research include Chen Wang,
Kaiyi Ji,
Junyi Geng,
Zhongqiang Ren,
Taimeng Fu,
Fan Yang,
Yifan Guo,
Haonan He,
Xiangyu Chen,
Zitong Zhan,
Qiwei Du,
Shaoshu Su,
Bowen Li,
Yuheng Qiu,
Yi Du,
Qihang Li,
Yifan Yang,
Xiao Lin,
Zhipeng Zhao.