Learned Grasping With Contact-Force Lookahead Selection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing robotics systems face challenges in determining optimal grasps for object manipulation, particularly in handling the complexity of contact forces and friction cones, which leads to slow runtime performance and the need for extensive training data.
Innovation Solution
The system employs a processor-based approach that identifies actions from a discrete action space, utilizes a transition model with contact parameters to generate a second set of grasps, and applies a reward function optimized using a lookahead tree to evaluate and select the next grasp, thereby enhancing online augmentation for learned grasping.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If existing robotics systems use traditional methods to determine optimal grasps considering contact forces and friction cones, then grasp accuracy is improved, but computation time increases significantly
Solution Approach 1:
The system performs preliminary training offline to learn contact force models and friction cone parameters from simulation data. During online operation, the pre-trained model rapidly evaluates grasps without重新 computing complex contact mechanics, thus maintaining accuracy while reducing runtime computation time
Solution Approach 2:
The system dynamically adjusts the level of computation based on operational context. During training phase, full physics simulations are used to build accurate models. During deployment, the system switches to faster approximate evaluations using learned models, optimizing the trade-off between accuracy and speed for each operational phase
2Reliability
If existing robotics systems use comprehensive training data for grasp learning, then grasp reliability is improved, but data requirements and training time increase
Solution Approach 1:
The system performs preliminary action by pre-training contact force models and friction cone parameters offline using simulation data before deployment. This preliminary training phase captures essential physics relationships, allowing the system to achieve high reliability during online operation with minimal additional training data requirements
Solution Approach 2:
The system uses simulation environments to create virtual copies of physical scenarios for training. By learning from simulated contact interactions and friction cone behaviors in a virtual workspace, the system acquires reliable grasp knowledge without requiring extensive real-world trial-and-error data collection
3Ease of manufacture
If the system uses a discrete action space for grasp selection, then implementation simplicity is improved, but grasp quality and adaptability deteriorate
Solution Approach 1:
The system changes parameters by representing continuous contact forces as discrete categorical variables during training, then using these learned discrete representations to guide continuous grasp selection during deployment. This parameter transformation allows the system to maintain implementation simplicity through discrete action spaces while achieving high grasp quality through learned force models
Data Source
AI summary
Systems and methods for online augmentation for learned grasping are provided. In one embodiment, a method is provided that includes identifying an action from a discrete action space. The method includes identifying a second set of grasps of the agent utilizing a transition model based on the action and at least one contact parameter. The at least one contact parameter defines allowed states of contact for the agent. The method includes applying a reward function to evaluate each grasp of the second set of grasps based on a set of contact forces within a friction cone that minimizes a difference between an actual net wrench on the object and a predetermined net wrench. The reward function is optimized online using a lookahead tree. The method includes selecting a next grasp from the second set. The method includes causing the agent to execute the next grasp.