September 27, 2021
To function effectively in urban settings, mobile robots and other autonomous systems should be able to move safely on sidewalks and avoid collisions with pedestrians or other obstacles. This is especially true for delivery robots or systems specifically programmed to patrol urban environments.
Researchers at the Georgia Institute of Technology and Stanford University recently developed AlienGo, a quadruped robot that can follow specific routes generated by public mapping services while staying on sidewalks and avoiding collisions with obstacles or humans. This robot, featured in a pre-published article on arXiv, is based on a new high-performance two-step learning framework for safe sidewalk navigation.
“As part of this project, we have developed an intelligent quadrupedal robot capable of navigating sidewalks in the real world,” Sehoon Ha, one of the researchers who conducted the study, told TechXplore. “Our work draws on two lines of existing work: autonomous driving and indoor robotic navigation. However, as navigation on exterior sidewalks typically takes place in unstructured environments with a wide variety of pedestrians and obstacles. without any guidance path, we also came up with a set of learning techniques and algorithms to solve these specific challenges. “
Initially, the team trained an artificial neural network to navigate simple sidewalk environments in simulations. This first algorithm, nicknamed “the expert”, was trained using a high speed salient world simulator and gained access to what is called the “privileged state” of the simulation.
Subsequently, this “expert” network transferred the behavior it learned to a “student” algorithm in a high fidelity simulation. Ultimately, this “student” network produced realistic sensor observations that looked like real-world sidewalk images.
“The ‘student’ uses a network of custom semantic features to generate abstractions which are then used to control the robot,” Maks Sorokin, another researcher involved in the study, told TechXplore. “This approach is based on our experience that the desired behavior is difficult to achieve using naïve end-to-end training, simply because the problem is far too difficult.”
Using the two-step learning framework they developed, Ha, Sorokin and their colleagues were able to arrive at an effective policy by using ‘inside’ information in simulation, then transferring the behaviors acquired by the framework to a real one. robot on all fours. When the team evaluated the frame, they found that it outperformed other leading models for sidewalk navigation. They then also tested their frame in a real environment, applying it to the AlienGo robot as it navigated the sidewalks of Atlanta.
“In addition to the performance gains resulting from using two-step learning with the abstract world, it was surprising how easy the transfer to the real world was with our data augmentation / curation,” said added Sorokin. “Given that during training the robot never saw real-world sidewalk images and given all of the real-world complexities, performance without any adaptation was noticeable to say the least. Our results could imply that a lot of recent work learning about robotics could be transferred to the real world and hopefully used in practice for the benefit of humanity. ”
In the future, the quadruped robot developed by this team of researchers could be used to perform various tasks, such as delivering packages or monitoring urban environments. Additionally, the framework they developed could be applied to other existing or emerging mobile robots to improve their ability to move on sidewalks.
“Although we have made a lot of progress in sim-to-real transfer for navigation, there are still many challenges ahead,” Sorokin said. “Some of the navigational challenges that we still have to overcome include crossing roads, dynamic obstacle management, and interacting with real-world objects and humans. However, our approach is not limited to navigation, it could potentially be applied in many robot applications, such as manipulation, locomotion and others. We are delighted to see its applications in adjacent research areas. “
Jie Tan, one of the researchers on the team, is currently working at Google, but the opinions expressed in this article do not represent the opinions of Google.
A framework for the navigation of indoor robots in humans
Learn to navigate sidewalks in outdoor environments. arXiv: 2109.05603 [cs.RO]. arxiv.org/abs/2109.05603
© 2021 Science X Network
Quote: Researchers are developing a new robot capable of efficiently navigating sidewalks in urban settings (2021, September 27) retrieved September 27, 2021 from https://techxplore.com/news/2021-09-robot-efficiently-sidewalks-urban -environments.html
This document is subject to copyright. Other than fair use for private study or research purposes, no part may be reproduced without written permission. The content is provided for information only.