A robot that can autonomously explore real-world environments

Robotics specialists have developed many advanced systems over the last decade, but most of these systems still require some degree of human supervision. Ideally, future robots should explore unknown environments autonomously and independently, continuously collecting data and learning from this data.

Researchers at Carnegie Mellon University recently created ALAN, a robotic agent that can autonomously explore unknown environments. This robot, featured in an article previously published in arXiv and to be presented at the International Conference on Robotics and Automation (ICRA 2023), successfully completed real-world tasks after a brief amount of exploratory testing.

“We have been interested in building an AI that learns by setting its own goals,” Russell Mendonca, one of the researchers who conducted the study, told Tech Xplore. “By not relying on humans for supervision or guidance, these agents can keep learning in new scenarios, driven by their own curiosity. This would allow for continued generalization to different domains and the discovery of increasingly complex behaviors.”

The robotics group at Carnegie Mellon University had already introduced some autonomous agents that could perform well in new tasks with little or no additional training, including a model trained to play the video game Mario and a system that could complete object manipulation tasks in various stages. . However, these systems have only been trained and tested in simulated environments.

The key goal of the team’s recent study was to create a framework that could be applied to physical robots in the world, improving their ability to explore their environment and complete new tasks. ALAN, the system they create, learns to explore its environment autonomously, without receiving rewards or guidance from human agents. Later, you can reuse what you learned in the past to tackle new tasks or problems.

“ALAN learns a world model in which to plan its actions and direct itself using environment-centric and agent-centric objectives,” Mendonca explained. “It also narrows the workspace to the area of ​​interest using pre-trained detectors out of the box. After scanning, the robot can unite the discovered abilities to perform single-stage or multi-stage tasks specified via target images.”

The researchers’ robot features a visual module that can estimate the movements of objects in its environment. This module then uses these estimates of how the objects have moved to maximize the change in the objects and encourage the robot to interact with these objects.

“This is an environment-centric signal, as it does not depend on the agent’s belief,” Mendonca said. “To improve its estimation of change in objects, ALAN needs to be curious about it. To do this, ALAN uses its learned model of the world to identify actions where it is uncertain about the expected change of the object, and then executes them on the world. real”. world. This agent-centric signal evolves as the robot sees more data.”

Previously proposed approaches to exploring autonomous robots required large amounts of training data. This significantly prevents or limits its implementation in real robots. In contrast, the learning approach proposed by Mendonca and colleagues allows the ALAN robot to continuously and autonomously learn to complete tasks while exploring its environment.

“We showed that ALAN can learn to manipulate objects with only about 100 trajectories in 1 to 2 hours in two different game kitchens, without any reward,” Mendonca said. “Therefore, using visual background knowledge can greatly increase the efficiency of robot learning. Extended versions of this system running 24/7 will be able to continuously acquire new skills.” useful with minimal human intervention in all domains, bringing us closer to general knowledge”. -intelligent purpose robots.”

In initial evaluations, the team’s robot performed remarkably well, being able to quickly learn to complete new manipulation tasks without any training or help from human agents. In the future, ALAN and the framework that supports it could pave the way for the creation of better-performing autonomous robotic systems for environmental exploration.

“Next, we want to study how to use other antecedents to help structure the robot’s behavior, such as videos of humans performing tasks and language descriptions,” Mendonca added. “Systems that can effectively harness this data will be better able to explore autonomously when operating in structured spaces. Additionally, we are interested in multi-robot systems that can combine their expertise to continuously learn.”

© 2023 Science X Network