Core Concepts
By estimating the object's direction of corrective rotation using the displacement patterns of tactile sensor's black dots, the robot can manipulate the object's pose to achieve stable placement across a variety of objects.
Abstract
The paper proposes a method for stable object placing using vision-based tactile sensors, such as GelSight, as an alternative to traditional Force/Torque (F/T) sensors.
The key insights are:
- The displacement patterns of the black dots on the tactile sensor can be categorized into two types corresponding to the roll and pitch directions of corrective rotation.
- The direction of corrective rotation can be estimated by calculating the Curl (for pitch) and Diff (for roll) features from the black dot displacements, without the need for F/T sensor measurements.
- By controlling the robot's end-effector speed to minimize these Curl and Diff features, the system can achieve stable object placement with high accuracy (less than 1-degree error) in nearly 100% of cases across 18 diverse objects.
- The proposed tactile-based method outperforms the baseline F/T sensor-based approach, which fails for objects with small support polygons due to sensor noise and cable tension issues.
The method is versatile, handling objects with asymmetrical shapes, textures, small support polygons, soft materials, and changing centers of gravity, demonstrating its potential as an effective alternative to F/T sensors for stable object placing tasks.
Stats
The robot arm generates a force of 5 N along the z-axis of the robot coordinate system to press the object against the desk.