Grasping and In-Hand Manipulation

Grasp Synthesis

Grasp synthesis involves formulating the problem of finding a suitable grasp for an object as a geometric optimization task, often utilizing analytic methods such as wrench space calculations to determine feasible grasps. Wrench space provides a mathematical representation of forces and moments that can be used to constrain the possible grasp configurations, allowing the system to identify orientations where the object is stable and secure. Furthermore, learning-based predictors are increasingly employed, leveraging datasets of observed grasps to train models capable of predicting optimal grasp poses based on object geometry and context.

These predictive models often incorporate features derived from 3D scans of the target object, allowing them to adapt quickly to new shapes and sizes. Datasets used for training these predictors typically contain a wide range of objects with corresponding grasp pose data, enabling the system to generalize its grasping capabilities beyond the specific examples it was trained on.

Control and Tactile

Effective control of in-hand manipulation relies heavily on force control, slip detection, and the fusion of tactile and visual information to provide robust feedback. Precise force control allows the robot to apply just enough pressure to maintain a stable grasp without damaging the object, while slip detection algorithms monitor for unexpected movements that could compromise the grip. Tactile-vision fusion combines data from both sensors to create a more complete understanding of the object's state and the forces involved in manipulation.

This integrated approach enables the system to react quickly and accurately to changes in the environment, ensuring stable and reliable grasping performance.

Examples

Example: Regrasping for Assembly demonstrates a common scenario where an initial grasp fails due to slippage or instability. The system first plans an initial grasp based on the object's geometry and desired pose, followed by executing this grasp with tactile cues providing real-time feedback. Finally, accuracy and success rates are validated to assess the overall effectiveness of the regrasp strategy.

This iterative process allows for dynamic adjustments to the grasping strategy, improving robustness in complex manipulation tasks.

Frequently asked questions

Try it live

Everything above runs in your browser — open Inverse Kinematics (FABRIK) and change the parameters while it is running. Nothing is installed, nothing is uploaded, the whole model lives in one tab.