PaperBot: Learning to Design Real-World Tools Using Paper

Ruoshi Liu,Junbang Liang,Sruthi Sudhakar,Huy Ha,Cheng Chi,Shuran Song,Carl Vondrick

2024-03-15

Abstract:Paper is a cheap, recyclable, and clean material that is often used to make practical tools. Traditional tool design either relies on simulation or physical analysis, which is often inaccurate and time-consuming. In this paper, we propose PaperBot, an approach that directly learns to design and use a tool in the real world using paper without human intervention. We demonstrated the effectiveness and efficiency of PaperBot on two tool design tasks: 1. learning to fold and throw paper airplanes for maximum travel distance 2. learning to cut paper into grippers that exert maximum gripping force. We present a self-supervised learning framework that learns to perform a sequence of folding, cutting, and dynamic manipulation actions in order to optimize the design and use of a tool. We deploy our system to a real-world two-arm robotic system to solve challenging design tasks that involve aerodynamics (paper airplane) and friction (paper gripper) that are impossible to simulate accurately.

Robotics

What problem does this paper attempt to address?

The problem this paper attempts to address is how to design practical tools using paper as a material. Specifically, the authors propose a framework called PaperBot, which can autonomously learn how to design and use paper tools and apply these tools directly in the real world. The paper focuses on two specific design tasks:

1. **Designing and folding paper airplanes**: The goal is to maximize the flight distance of the paper airplane through a series of folding and throwing actions.
2. **Designing a paper gripper**: The goal is to maximize the gripping force by cutting and folding the paper.

### Main Issues
- **Limitations of traditional methods**: Traditional tool design methods often rely on simulation or physical analysis, which are usually inaccurate and time-consuming. Especially for paper, a thin-shell, easily deformable material, its complex mechanical behavior is difficult to simulate accurately.
- **Automated tool design**: How to use machine learning methods to automate the iterative process of design, deployment, and improvement, thereby efficiently designing optimized paper tools in the real world.

### Solution
- **Self-supervised learning framework**: PaperBot uses a self-supervised learning framework, combined with a robotic arm and related equipment (such as a Cricut cutting machine), to experiment and learn in the real world.
- **Optimizing design parameters**: Using neural networks as surrogate models to learn the nonlinear relationship between design parameters and performance, and optimizing design parameters through methods such as gradient descent.
- **High adaptability**: The system can quickly adapt to different task definitions without the need for manual customization of each design.

### Experimental Results
- **Paper airplane task**: After 100 trials (approximately 3 hours), the paper airplanes designed by PaperBot significantly outperformed other baseline methods, including state-of-the-art evolutionary algorithms, in terms of flight distance.
- **Paper gripper task**: Similarly, after 100 trials, the paper grippers designed by PaperBot could exert a maximum gripping force significantly greater than other baseline methods, equivalent to the weight of more than 4 strawberries.

### Contributions
- **Real-world tool design**: Proposes a method for directly designing and optimizing paper tools in the real world without relying on simulation and prior knowledge.
- **Efficient learning algorithm**: Through a self-supervised learning and optimization framework, it can find optimized designs within a limited number of trials.
- **Wide applicability**: Demonstrates the effectiveness and adaptability of the method in different tasks and design spaces, providing new ideas for future tool design.