IT House June 8 news, recently, some researchers with the help of ChatGPT to create a tomato picking robot.
OpenAI's ChatGPT is a language model capable of processing and understanding large amounts of text data and using that information to answer questions. In a new case study, researchers from Delft University of Technology in the Netherlands and the Swiss Federal Institute of Technology turned to ChatGPT-3 for help in the design and manufacture of the robot.
This sounds a bit strange because ChatGPT-3 is a language model, not a mechanical model. "Although ChatGPT-3 is a language model, its code generation is also text-based, providing important insights and intuition into physical design and showing great potential to unleash human creativity," said Josie Hughes, co-author of the study.
First, the researchers asked the AI model, "What are the challenges facing humanity in the future?" "ChatGPT-3 proposes three: food supply, aging population, and climate change. The researchers chose food supply as the direction of the robot's design and then asked ChatGPT-3 what characteristics a tomato harvesting robot should have. ChatGPT-3 proposes a motor-driven clip for pulling ripe tomatoes from vines.
Once this rough design is determined, the researchers can make decisions about design details, including what materials to use and what control code to write. Currently, language models cannot generate complete computer-aided design (CAD) models, evaluate code, or automate the manufacturing of robots, so this step requires researchers to act as "technicians" to assist with these aspects, optimize the code written by language models, complete the CAD, and build the robot.
Illustration: An overview of the discussion between researchers and language models, with questions asked by humans above and options offered by language models below. Green shades indicate a human decision tree, gradually focusing the problem in a direction that aligns with their goals.
Based on technical advice provided by ChatGPT-3, the researchers made their robotic clip and tested it in the real world, using it to pick tomatoes, and the results were successful.
a. Some technical recommendations generated by the language model, including shape indication, code, component and material selection, mechanical design, etc. Guided by these inputs, a clip is manufactured and tested on real-world tasks such as tomato picking, as shown in the image on the right.
The researchers say their case study demonstrates the potential of collaboration between humans and language models to transform the design process, but they also realize that this collaboration may have varying degrees of work.
In one extreme way, they say, AI acts as the "inventor," providing all the input to the robot's design, while humans simply apply it blindly. Another way is to use AI's extensive knowledge to supplement human expertise. The third way is to retain humans as inventors, using AI to troubleshoot, debug, and handle tedious or time-consuming processes to optimize the design process.
The researchers raised some ethical and commonsense risks that could arise as a result of AI collaborating with humans, pointing to issues such as bias, plagiarism, and intellectual property as areas to focus on.
"In our study, ChatGPT-3 concluded that tomatoes were the most worthwhile crop for which a harvesting robot was designed," Hughes said, "but this may be biased against crops with more literature coverage than those that are really needed." When decisions are made beyond the scope of an engineer's knowledge, this can lead to serious ethical, engineering, or factual errors. ”
Despite these problems, the researchers believe that collaboration between AI and humans has great potential if managed properly.