Grasp Parameter Determination via Kinesthetic Teaching
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Robots face challenges in determining appropriate grasp parameters for objects, often failing to leverage physical manipulation data or object models, leading to inefficient grasping techniques.
Innovation Solution
The method involves determining grasp parameters based on sensor data generated during physical manipulation by a user, including end effector poses and translational force measures, which are then associated with an object model to guide robotic actuators in grasping objects effectively.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If robots use traditional programming methods to grasp objects, then they can perform basic grasping tasks, but they fail to leverage physical manipulation data and object models, resulting in inefficient and inflexible grasping techniques
Solution Approach 1:
The system performs preliminary actions by collecting sensor data during user physical manipulation of objects and pre-processing this data to extract grasp parameters. This preliminary data collection and processing enables the robot to learn effective grasping techniques before actual autonomous operation, improving both adaptability and efficiency.
Solution Approach 2:
The system creates a digital copy of the physical manipulation process by recording sensor data (forces, torques, positions) during user handling of objects. This copied data is then stored in object models and reused to guide robot grasping actions, eliminating the need for repeated trial-and-error programming while maintaining high adaptability.
2Measurement precision
If robots collect and process sensor data from physical manipulation, then they can determine accurate grasp parameters, but this requires complex data processing and association with object models
Solution Approach 1:
The system merges multiple data sources (force sensors, torque sensors, position sensors) into a unified object model that associates all grasp parameters with a single object representation. This consolidation simplifies data processing by providing a centralized structure rather than requiring separate processing for each sensor type.
Solution Approach 2:
The object model serves as an intermediary structure that mediates between raw sensor data and robot control commands. Instead of directly processing complex sensor data for each grasping task, the system uses the pre-built object model as an intermediate layer that translates sensor observations into actionable grasp parameters, reducing processing complexity.
3Force
If robots use impactive or ingressive end effectors to grasp objects, then they can physically manipulate objects, but they may apply excessive force or fail to account for object fragility
Solution Approach 1:
The system uses force and torque sensors to provide continuous feedback during object manipulation. By monitoring the forces applied during user physical manipulation and storing this data in object models, the robot learns the appropriate force levels for different objects, enabling it to adjust its grasping force to avoid damage while maintaining effective manipulation.
Data Source
AI summary
Some implementations are directed to methods and apparatus for determining, based on sensor data generated during physical manipulation of a robot by a user, one or more grasp parameters to associate with an object model. Some implementations are directed to methods and apparatus for determining control commands to provide to actuator(s) of a robot to attempt a grasp of an object, where those control commands are determined based on grasp parameters associated with an object model that conforms to the object. The grasp parameter(s) associated with an object model may include end effector pose(s) that each define a pose of a grasping end effector relative to the object model and/or translational force measure(s) that each indicate force applied to an object by a grasping end effector, where the force is at least partially the result of translation of an entirety of the grasping end effector.