Andrew S. Morgan,Bowen Wen,Junchi Liang,Abdeslam Boularias,Aaron M. Dollar,Kostas Bekris
Abstract:Highly constrained manipulation tasks continue to be challenging for autonomous robots as they require high levels of precision, typically less than 1mm, which is often incompatible with what can be achieved by traditional perception systems. This paper demonstrates that the combination of state-of-the-art object tracking with passively adaptive mechanical hardware can be leveraged to complete precision manipulation tasks with tight, industrially-relevant tolerances (0.25mm). The proposed control method closes the loop through vision by tracking the relative 6D pose of objects in the relevant workspace. It adjusts the control reference of both the compliant manipulator and the hand to complete object insertion tasks via within-hand manipulation. Contrary to previous efforts for insertion, our method does not require expensive force sensors, precision manipulators, or time-consuming, online learning, which is data hungry. Instead, this effort leverages mechanical compliance and utilizes an object agnostic manipulation model of the hand learned offline, off-the-shelf motion planning, and an RGBD-based object tracker trained solely with synthetic data. These features allow the proposed system to easily generalize and transfer to new tasks and environments. This paper describes in detail the system components and showcases its efficacy with extensive experiments involving tight tolerance peg-in-hole insertion tasks of various geometries as well as open-world constrained placement tasks.
Robotics,Artificial Intelligence,Computer Vision and Pattern Recognition,Systems and Control
What problem does this paper attempt to address?
The problem that this paper attempts to solve is how autonomous robots can achieve reliable and high - precision operations in high - precision assembly tasks under strict constraints (such as tolerances less than 1 millimeter). Specifically, the paper focuses on how to complete these tasks requiring extremely high precision through vision - driven compliant manipulation, combined with advanced object - tracking techniques and passively - adaptive mechanical hardware. Traditional methods often rely on expensive force sensors, precise manipulators or time - consuming online learning, and these methods require a large amount of data. In contrast, the method proposed in this paper does not require these expensive devices or complex online learning processes. Instead, it utilizes mechanical compliance and an off - line - learned hand - manipulation model, as well as RGBD - based object - tracking techniques, to achieve accurate object - insertion tasks.
### Main contributions of the paper
1. **No need for force sensors or precise manipulators**: By taking advantage of mechanical compliance, the system can complete high - precision insertion tasks without force sensors or precise manipulators.
2. **Improve insertion reliability**: By performing within - hand manipulation, the workspace is expanded, thereby increasing the success rate of insertion tasks.
3. **The role of visual feedback**: Demonstrates the importance of visual feedback in high - precision tasks, especially that it can handle insertion tasks more effectively under the action of system compliance.
### Specific methods
- **Visual object pose tracking**: Use an advanced low - latency RGBD object pose - tracking technique. This technique is trained only with synthetic data and can accurately estimate the 6D pose of an object and remains robust even under occlusion and different lighting conditions.
- **Object - independent within - hand manipulation**: Use a learned inverse system - dynamics model that is object - independent and can be used for within - hand object manipulation to align the object orientation for easy insertion.
- **Visual - feedback controller**: Develop an insertion - control strategy that relies entirely on the feedback of the task state, closes the control loop through the 6D pose estimation of the object tracker, and can be applied to objects of different geometries, thus being suitable for different scenarios.
### Experimental verification
The paper verifies the effectiveness of the proposed system through a large number of experiments, including tight - tolerance pin - hole insertion tasks of different geometries, and open - world constrained - placement tasks, such as cup - stacking, plug - insertion into sockets, marker - insertion into brackets and box - packing, etc., demonstrating the practical application value of the system in complex operation scenarios.
In conclusion, this paper proposes an innovative vision - driven compliant - manipulation framework that can achieve high - precision assembly tasks without expensive sensors and complex online learning, and has broad application prospects.