TERP: a method for achieving reliable robotic navigation in rough outdoor terrain

Trajectories of the robot when navigating in rough terrain (altitude gain ≥ 3 m) using four different methods: our TERP method (pink), Ego-graph (orange), DWA (blue) and the Attn-DRL method of end to end (purple) . TERP generates waypoints (pink dots) that are dynamically feasible and accessible by local paths at lower cost. Other methods navigate in dangerous regions with high elevation gradients that could lead to unstable robot orientations. The TERP leads to trajectories with low elevation gradients with a higher success rate of reaching objectives. Credit: Weerakoon et al.

Autonomous mobile robots are already being tested and used for applications such as package delivery, surveillance, search and rescue missions, planetary/space exploration and environmental monitoring. For these robots to successfully accomplish their missions, they must be able to operate safely and reliably in rough outdoor terrain, without colliding with nearby obstacles.

Researchers at the University of Maryland, College Park (UMDCP) recently developed a new machine learning method that could improve the reliability of robot navigation in rough outdoor terrain and in the presence of obstacles. Their study was presented by the GAMMA research group of the UMDCP at the IEEE International Conference on Robotics and Automation 2022.

“We observed that geometric features of the terrain, including changes in elevation or roughness, significantly affect the stability of a robot’s movements during navigation,” Dinesh Manocha, a professor at the UMDCP who led this research project. “Therefore, it is essential for robots to perceive these terrain features in the environment to make safe navigation decisions. “

Before Manocha and his students began working on their robotic navigation method, researchers closely observed the locomotion strategies of humans as they move through a complex outdoor environment. Interestingly, they noticed that humans did not focus on the whole environment when moving, but rather on regions of space that they considered critical or important.

The robot navigation method developed by Manocha’s group was inspired by the locomotion behavior they observed in humans. This method, called TERP (Terrain Elevation-based Robot Path planning), is based on a deep reinforcement learning (DRL) technique they developed.

“Our new hybrid machine learning architecture combines the intermediate output results of our attention-based DRL network with a novel path planning method,” said Weerakoon, a Ph.D. student working on the project. Explain. “These intermediate results help to identify and avoid difficult or dangerous areas in the environment during navigation. Our approach uses a fully trained DRL network that uses elevation maps, the robot’s pose, and its target as inputs to calculate a mask of attention.

The attention mask calculated by the team’s algorithm then guides a mobile robot to regions of its surrounding environment that it needs to pay special attention to achieve stable navigation. Ultimately, this mask is combined with the input elevation map produced by the process, creating a 2D navigation cost map. This map is then used to plot a safe and reliable path for the robot to reach a desired location.

Trajectories of the robot while navigating different simulated and real uneven terrain using TERP (pink), TERP without attention (yellow), end-to-end Attn-DRL network (purple), ego-graph (orange), Egograph+ (green) and DWA (blue). (a) High altitude; (b) Citycurb; (c) Low Altitude; (d) Mid Altitude; (e) real-world average elevation; (f) real world edge; (g) real world average elevation; (h) real-world average elevation with obstacle regions; We observe that TERP calculates trajectories with low elevation gradients over rough terrain and is able to handle difficult sidewalk scenarios. Credit: Weerakoon et al.

“In previous work, we observed significant performance degradation in end-to-end DRL methods when transferring from simulation to real-world terrains,” said Sathyamoorthy, another Ph.D. student working on this project. , said. “However, our new hybrid machine learning architecture improves navigation performance.”

The attention component of the TERP can dramatically improve a robot’s spatial awareness, simply by shifting its attention to the regions most critical for the navigation task at hand. On the other hand, the waypoint planning component of their method ensures that the robot follows the most cost-effective path to reach its destination.

“TERP generates relatively stable trajectories in steep elevations to minimize the risk of robot rollover,” said Patel, a research staff member working on the project. “In addition, it can avoid dangerous regions and obstacles when navigating complex terrain with static and dynamic obstacles.”

Manocha and his students evaluated their method in different real-world environments, using the Husky robot, an unmanned mobile robotic system developed by Clearpath Robotics. In their tests, the robots navigated outdoor spaces with rough terrain, with an elevation gain of up to 4 meters.

“We have shown that our unique hybrid formulation with an attention DRL network for perception and a waypoint planner for navigation leads to a high navigation success rate over complex terrain,” Manocha said. “This implies that our method significantly reduces the risk of the robot rolling over when navigating difficult rough terrain.”

In the team’s initial evaluations, TERP achieved remarkable results, suggesting that it can significantly improve the robot’s navigation reliability in complex outdoor environments. In the future, it could be used to improve the performance of robots in many contexts, for example opening up new possibilities for planetary and space explorations, agricultural surveys and complex environmental monitoring.

“We plan to improve our system in the future by addressing some of its current limitations,” Manocha added. “In particular, we have observed that in addition to terrain elevation, surface properties such as texture, bumps and deformability govern a robot’s navigability in complex outdoor scenarios and we are working on methods self-supervised learning methods to handle such scenarios.We also extend these methods for the autonomous navigation of legged robots, such as the Boston Dynamics Spot robot.


A beaver-inspired method to guide the movements of a one-legged swimming robot


More information:

TERP: Reliable planning in uneven outdoor environments using deep reinforcement learning. GAMMA Group, University of Maryland, College Park (2022). arXiv:2109.05120 [cs.RO] arxiv.org/abs/2109.05120

© 2022 Science X Network

Quote:
TERP: A method to achieve reliable robotic navigation in rough outdoor terrain (June 2, 2022)
retrieved 2 June 2022
from https://techxplore.com/news/2022-06-terp-method-reliable-robot-uneven.html

This document is subject to copyright. Apart from any fair use for the purpose of private study or research, no
any part may be reproduced without written permission. The content is provided for information only.

Stay connected with us on social media platform for instant update, click here to join our Jwitter& Facebook


Source link

About Donald P. Hooten

Check Also

The 369 Manifestations Method: How to Make Your Therapy Sessions Work

Psychologist Nancy Sokarno walks us through the 369 Manifestations Method and how to get the …