Words prove their value as educational tools for robots – Artificial intelligence and robotics news

Exploring a new way to teach robots, Princeton researchers found that human-language descriptions of tools can speed up learning of a simulated robotic arm lift and the use of a variety of tools. .

The findings build on evidence that providing richer information during artificial intelligence (AI) training can make autonomous robots more adaptable to new situations, improving their safety and efficiency.

Adding descriptions of a tool’s form and function to the robot’s training process improved the robot’s ability to handle newly encountered tools that were not in the original training set. A team of mechanical engineers and computer scientists presented the new method, Accelerated Learning of Tool Manipulation with LANguage, or ATLA, at the Robot Learning Conference on December 14.

Robotic arms have great potential to help perform repetitive or difficult tasks, but it’s difficult to train robots to handle tools effectively: tools come in a wide variety of shapes, and a robot’s dexterity and vision are no match for those of a human.

“Additional information in the form of language can help a robot learn to use tools faster,” said study co-author Anirudha Majumdar, an assistant professor of mechanical and aerospace engineering at Princeton who leads the Intelligent RobotMotionLab.

The team obtained tool descriptions by querying GPT-3, a large language model released by OpenAI in 2020 that uses a form of AI called deep learning to generate text in response to a prompt. After experimenting with various prompts, they decided to use “Describe the [feature] of [tool] in a detailed and scientific answer”, where the characteristic was the shape or purpose of the tool.

“Because these language models were trained on the Internet, in a sense you can think of that as a different way of retrieving this information,” more efficiently and comprehensively than using crowdsourcing or scraping sites Web sites for tool descriptions, said Karthik Narasimhan, an assistant professor of computer science and co-author of the study. Narasimhan is a core faculty member in Princeton’s Natural Language Processing (NLP) group and contributed to the original GPT language model as a visiting scholar at OpenAI.

This work is the first collaboration between the Narasimhan and Majumdar research groups. Majumdar is focused on developing AI-based policies to help robots – including flying and walking robots – generalize their functions to new settings, and he was curious about the potential of recent “massive advances in the treatment of natural language” for the benefit of robot learning, he told me.

For their simulated robot learning experiments, the team selected a training set of 27 tools, ranging from an ax to a squeegee. They gave the robotic arm four different tasks: pushing the tool, lifting the tool, using it to sweep a cylinder along a table, or driving a peg into a hole. The researchers developed a suite of policies using machine learning training approaches with and without linguistic information, then compared policy performance on a separate test set of nine tools with matched descriptions.

This approach is known as meta-learning, since the robot improves its ability to learn with each successive task. It’s not just about learning how to use each tool, but also “trying to learn to understand the descriptions of each of those hundred different tools, so when he sees the 101st tool, he’s quicker to learn how to use the new tool,” Narasimhan said. “We do two things: we teach the robot how to use the tools, but we also teach it English. »

The researchers measured the robot’s success at pushing, lifting, sweeping, and pounding with the nine test tools, comparing results with policies that used language in the machine learning process to those that didn’t. linguistic information. In most cases, the linguistic information provided significant benefits for the robot’s ability to use new tools.

One task that showed notable differences between policies involved using a crowbar to sweep a cylinder or bottle across a table, said Allen Z. Ren, a Ph.D. student in Majumdar’s group and main author of the research paper.

“With language training, he learns to grab the long end of the crowbar and use the curved surface to better constrain the movement of the bottle,” Ren said. “Without the tongue, he grabbed the crowbar near the curved surface and it was harder to control. »

The research was supported in part by the Toyota Research Institute (TRI) and is part of a larger TRI-funded project in Majumdar’s research group aimed at improving the ability of robots to operate in new situations that differ of their training environments.

“The overall goal is to make robotic systems — especially those trained using machine learning — generalize to new environments,” Majumdar said. Other TRI-supported work by his group has focused on failure prediction for vision-based robot control and used an “adversarial environment generation” approach to help robot policies perform better in conditions outside of their initial training.

The item, Leverage language for accelerated learning of tool manipulation, was presented on December 14 at the Robot Learning Conference. Besides Majumdar, Narasimhan and Ren, co-authors include Bharat Govil, Princeton Class of 2022, and Tsung-Yen Yang, who earned a Ph.D. in Electrical Engineering at Princeton this year and is now a Machine Learning Scientist at Meta Platforms Inc.

In addition to TRI, research support was provided by the US National Science Foundation, the Office of Naval Research, and the School of Engineering and Applied Science at Princeton University through the generosity of William Addy ’82.

We would like to say thanks to the author of this short article for this amazing web content

Words prove their value as educational tools for robots – Artificial intelligence and robotics news

You can find our social media profiles , as well as other related pageshttps://yaroos.com/related-pages/