banner

Blog

Nov 06, 2024

ChatGPT trains $120 robot arm to master spill cleanup in 4 days

Each motion policy trained on 100 demos, helping robots learn from human actions, boosting efficiency and intuitive task handling.

Jijo Malayil

In just four days, the team developed a visual language model to enable smooth human-robot interaction.

In a remarkable demonstration of open-source innovation, a team has developed a cheap cleaning robot to complete a set of actions.

A pair of roboticists at UC Berkeley and ETH Zurich successfully trained a duo of $120 robot arms to handle spill cleanup.

In just four days, they programmed the robots using GPT-4o, creating a visual language model for effective human-robot interaction (HRI).

Their training involved approximately 100 demonstrations to refine the robots’ movements, showcasing the potential of accessible technology in automating everyday tasks.

Jannik Grothusen and Kaspar Janssen’s innovative project features the SO-100 robot arms developed by The Robot Studio, designed specifically for tasks like cleaning spills.

These robot arms are controlled by a sophisticated multimodal agent orchestrated using LangChain AI, which enables seamless communication and coordination between the robots and their environment.

The integration of this technology allows for enhanced human-robot interaction, making it easier for users to understand and engage with the robots during their operation.

To refine the robot arms’ movements, the team implemented motion policies based on LeRobot’s advanced implementation of the ACT (Actor-Critic Training) framework. This approach ensures that the robots learn effective motion patterns through reinforcement learning, allowing them to adapt and improve over time.

Each motion policy was trained using approximately 100 demonstrations, providing the robots with a robust dataset to learn from. This training method not only enhances the robots’ efficiency in handling various tasks but also contributes to their ability to learn from human actions, leading to more intuitive interactions.

The project exemplifies robotics’ growing accessibility and potential, showcasing how open-source resources and collaborative platforms can empower innovators to create advanced automation solutions.

If you’re looking for a budget-friendly robot for similar DIY projects at home, The Robot Studio has shared plans on YouTube for building the robot arms.

Their SO-100, used here, is an open-source 3D-printed robot arm designed for affordability and ease of assembly. It requires only six servos, one adapter board, and 3D-printed parts. All necessary files are available on Robot Studio GitHub, costing under $100 per unit with bulk purchase.

The arm’s parts, printed using durable carbon fiber material, are optimized for strength and easy detachment. With adjustable tolerances and a refined jaw design, the project aims to support ongoing customization and user-friendliness.

The assembly process requires setting servo support to 45° for optimal operation. Each servo receives a unique ID number, programmed through an adapter board connected to power. The servos are daisy-chained and identified in sequence.

Screws are tightened for stability with servo horns oriented either vertically or horizontally. Using a magnetic screwdriver for convenience, the rotation pitch is the first step in arm assembly. In order to provide unfettered rotation and a smooth, functional finish, screws are adjusted to “finger tight” for balanced movement.A servo-driven robotic arm is simple but accurate to assemble. Wires must be untangled and free of pinch points before important parts, such as servos and passive spools, are carefully positioned and fastened. Fingertight screws are used to mount each servo, ensuring that connections are secure without being overtightened.Magnetic screwdrivers and careful alignment maintain a smooth fit. Wrist pitch and roll are among the critical joints positioned to permit the best possible movement. Last-minute changes, such as modifying signal temporal logic (STL) files, improve the design’s functionality and longevity. In just 30 minutes, the end effect is a powerful, swift arm.

Stay up-to-date on engineering, tech, space, and science news with The Blueprint.

By clicking sign up, you confirm that you accept this site's Terms of Use and Privacy Policy

Jijo Malayil Jijo is an automotive and business journalist based in India. Armed with a BA in History (Honors) from St. Stephen's College, Delhi University, and a PG diploma in Journalism from the Indian Institute of Mass Communication, Delhi, he has worked for news agencies, national newspapers, and automotive magazines. In his spare time, he likes to go off-roading, engage in political discourse, travel, and teach languages.

a day ago

a day ago

a day ago

a day ago

Jijo Malayil
SHARE