Dynamic Catching Based on High Speed Visual Information

Mitsuru Higashimori, Masayuki Kimura, Akio Namiki, Mitsuhiko Ishikawa, Masahiro Kaneko, Idaku IshiiThe Proceedings of JSME Annual Conference on Robotics and Mechatronics (Robomec) 2004 · DOI

Abstract

This paper details the development of a robotic system capable of "Dynamic Catching"—the high-speed interception and grasping of fast-moving or fragile objects in mid-air. By integrating ultra-high-speed vision with a specialized three-fingered robotic hand featuring high-torque actuators and soft gel fingertips, the system achieves a millisecond-scale perception-action loop. Experimental results demonstrate that the synergy of high-speed control and mechanical compliance allows the robot to catch falling objects, including fragile eggs, without breakage by dynamically adjusting its posture to minimize impact forces.

Introduction

Catching a falling or thrown object is a task humans perform with ease but remains a significant challenge in robotics. Successful catching requires a "tight loop" between perception and action. Standard robotic systems typically fail in dynamic environments due to two primary factors: latency and impact dynamics. Standard vision sensors (30–60 fps) and conventional controllers introduce too much delay to track high-velocity objects, while rigid robotic fingers generate high impulse forces upon contact, often causing objects to bounce away or sustain damage.

This research, originating from the University of Tokyo and Hiroshima University, addresses these gaps by treating catching as an "active" dynamic interaction rather than a static "wait and grab" task. The authors propose a system that utilizes high-speed hardware and adaptive control strategies to manage the physics of impact in real-time.

Methodology

High-Speed Hardware and Vision

The core of the system is a three-fingered, eight-joint robotic hand designed for extreme acceleration. The hand utilizes high-power motors paired with small harmonic gears, enabling joints to rotate 180 degrees in just 0.1 seconds. Despite this speed, the fingertips can exert up to 28N of force, ensuring a secure grasp post-impact.

To guide this hardware, the system integrates a high-speed vision sensor capable of tracking objects at frame rates significantly higher than industrial standards (typically 1000 fps). This provides the real-time spatial coordinates necessary for the "Active Catching" algorithm to function.

Active Catching Strategy

Unlike "passive" catching, where a robot assumes a fixed posture and waits for contact, the "Active Catching" strategy uses continuous visual feedback to adjust the hand's position and orientation up to the millisecond of impact. This allows the controller to match the object's velocity or find the optimal geometric contact point instantaneously, effectively dissipating kinetic energy.

Soft Robotics Integration

To handle fragile objects such as eggs, the researchers equipped the fingers with soft gel fingertips. This mechanical impact absorption works in tandem with the high-speed control loop. While the gel handles the high-frequency physics of the initial collision, the high-speed actuators perform the macro-adjustments required to secure the object without exceeding its structural limits.

Experimental Results

The system was evaluated through physical trials involving falling cylindrical objects and fragile eggs. Key performance metrics included:

  • Actuation Speed: The system demonstrated joint speeds of 180°/0.1s, vastly outperforming standard industrial grippers.
  • Force Capability: A fingertip force of 28N was achieved, sufficient for stabilizing objects after high-speed interception.
  • Catching Success: The robot successfully caught falling eggs without breakage, validating the efficacy of combining soft fingertips with high-speed visual modulation.

Discussion and Insights

A primary insight of this work is the "Soft + Fast" synergy. Softness alone is insufficient for catching fragile objects at high speeds, as is speed without compliance. The research also explores bio-mimetic control, specifically methods that imitate the human nervous system's ability to suppress reflexes during voluntary movement. This allows for a more dexterous transition from a high-speed approach to a delicate, secure grasp.

Conclusion and Future Work

The current system successfully demonstrates active catching for objects on relatively simple vertical trajectories. However, challenges remain regarding "preshaping" for complex, rotating objects and handling parabolic paths with significant spin. Future research will focus on integrating more sophisticated grasping timing logic and miniaturizing the high-torque actuation for use in mobile or collaborative robotic platforms.

Summary updated

BibTeX
@inproceedings{higashimori2004dynamic,
  author = {Mitsuru Higashimori and Masayuki Kimura and Akio Namiki and Mitsuhiko Ishikawa and Masahiro Kaneko and Idaku Ishii},
  title = {{Dynamic Catching Based on High Speed Visual Information}},
  booktitle = {The Proceedings of JSME Annual Conference on Robotics and Mechatronics (Robomec)},
  year = {2004},
  doi = {10.1299/jsmermd.2004.108\_5},
}
Figure