Motion planning algorithms that can think like humans

New Delhi: Researchers at Indian Institute of Technology (IIT), Madras have developed a class of fast and efficient 'motion planning' algorithms which can think like human beings and enable autonomous aerial, ground or surface vehicles to navigate obstacle-cluttered environments. According to the team, the algorithms have been developed on a novel notion of 'Generalised Shape Expansion' (GSE) that enables planning for a safe and dynamically feasible trajectory for autonomous vehicles.

These approaches have been found to yield superior results compared to many of the existing seminal and state-of-the-art motion planning algorithms. Because of its novel calculation of 'safe' region, it provides a crucial advance during time-sensitive planning scenarios arising in applications like self-driving cars, disaster response, ISR operations, aerial drone delivery and planetary exploration, among others, the team claimed. The research led by Satadal Ghosh, Assistant Professor, Department of Aerospace Engineering, IIT Madras, has published several research papers in internationally reputed peer-reviewed journals like AIAA Journal of Guidance, Control, and Dynamics, and IEEE Control Systems Letters, and top-tier conferences like IEEE Conference on Decision and Control (CDC), American Control Conference (ACC) and AIAA SciTech.

The team included IIT Madras alumni Vrushabh Zinage who is currently a doctoral research scholar at University of Texas Austin (USA), Adhvaith Ramkumar who is currently a graduate student at Warsaw University of Technology, Poland and Nikhil P, an analyst at Goldman Sachs. "The GSE-based algorithms function by calculating a 'safe' region consisting of large 'visible' areas in the environment, customised to ensure navigability. Following this, the algorithms select a random point in this 'visible' region and connect it through a safe 'edge' to the safely reachable regions discovered so far. Eventually, the algorithms can almost always connect any two points in any environment, which satisfies certain basic criteria," Zinage said.

The researcher explained that the GSE-based algorithms' main advantage lies in the significant improvement of computational efficiency over several other well-established motion planning algorithms. This naturally leads to strong applicability of the GSE-based algorithms in applications, where planning is time-sensitive. "Instead of using computationally heavy dedicated collision checking modules, these algorithms leverage the novel notion of 'generalised shape', which gives a maximal representation of the free-space that is reachable from a point in the environment, which is almost similar to updating of human perception about the 'safe' space to move through surrounding him or her," said Adhvaith Rajkumar.