# Dynamically Constrained Motion Planning Networks for Non-Holonomic Robots

**Authors:** Jacob J. Johnson, Linjun Li, Fei Liu, Ahmed H. Qureshi, Michael C. Yip  
**arXiv:** [2008.05112](https://arxiv.org/abs/2008.05112)  
**Date:** 2020-08-12  
**Summary page:** https://www.summarizepaper.com/en/arxiv-id/2008.05112v1/

## Key points

- Dynamic Motion Planning Networks (Dynamic MPNet) is an algorithm for real-time motion planning in kinematically constrained robots.
- Real-time planning is crucial in today's automated ecosystem.
- Traditional methods that relax constraints often become unreliable or slow for these types of robots.
- The proposed algorithm utilizes a neural planning approach and modifies the training and planning networks of Dynamic MPNet.
- These modifications enable real-time planning while improving data efficiency and generalizability of trained models.
- The algorithm is evaluated through simulations for non-holonomic robot planning tasks and experimental results are demonstrated for indoor navigation using a Dubins car.
- This research provides a reliable and efficient solution for real-time motion planning in kinematically constrained robots.
- By incorporating neural networks, traditional methods are enhanced, resulting in improved overall performance.
- This work has potential applications in various industries where robots play a significant role in automation processes.

## Summary

This paper presents the algorithm Dynamic Motion Planning Networks (Dynamic MPNet), an extension of Motion Planning Networks for real-time motion planning in kinematically constrained robots. Real-time planning is crucial in today's rapidly expanding automated ecosystem. Traditional methods that relax constraints often become unreliable or slow for these types of robots. The proposed algorithm addresses this challenge by utilizing a neural planning approach and making modifications to the training and planning networks of Dynamic MPNet. These modifications enable real-time planning while also improving data efficiency and generalizability of trained models. The algorithm is evaluated through simulations for planning tasks in a non-holonomic robot and experimental results are demonstrated for an indoor navigation task using a Dubins car. This research contributes to the field of robotics by providing a reliable and efficient solution for real-time motion planning in kinematically constrained robots. By incorporating neural networks, traditional methods are enhanced, resulting in improved overall performance. This work has potential applications in various industries where robots play a significant role in automation processes.