In-Hand Manipulation Re-Planning With Grasp Feedback
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for dexterous in-hand manipulation lack robustness and fail to incorporate feedback, especially when dealing with parametric uncertainty in object parameters, and are not easily extendable to scenarios requiring online iterative re-planning.
Innovation Solution
An online iterative re-planning method that uses learned cost models and signed-distance functions to select contact points on objects, incorporating feedback from actual grasps and employing a hybrid approach of sampling and gradient-based optimization to plan sequences of object poses and grasps, enhancing robustness and failure recovery.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If existing methods for dexterous in-hand manipulation are used, then the basic manipulation task can be performed, but the system lacks robustness and fails to incorporate feedback from actual grasps
Solution Approach 1:
The patent implements feedback incorporation by using the actual achieved grasp and object pose from execution as inputs for generating new candidate trajectories. The system receives feedback from the actual grasp outcome and uses it to iteratively re-plan, allowing the system to adapt to parametric uncertainty and improve robustness through continuous refinement based on real execution results.
2Reliability
If existing methods are used, then simple manipulation can be achieved, but they fail when dealing with parametric uncertainty in object parameters
Solution Approach 1:
The patent applies dynamics by implementing an iterative re-planning system that continuously adapts to new information. Instead of using a static planning approach, the system dynamically generates new candidate trajectories based on actual achieved grasps and poses, allowing it to handle parametric uncertainty by updating its plan according to real execution outcomes.
Solution Approach 2:
The patent generates multiple candidate object trajectories in advance before execution, evaluating them using a learned cost model. This preliminary generation and evaluation of multiple possibilities allows the system to prepare for parametric uncertainty by having pre-computed alternative plans that can be selected based on actual execution feedback.
3Adaptability or versatility
If traditional planning methods are used, then computation time may be reduced, but the system is not easily extendable to scenarios requiring online iterative re-planning
Solution Approach 1:
The patent segments the planning process into distinct phases: generating candidate trajectories, evaluating them with a learned cost model, selecting the best candidate, and executing. This segmentation allows the system to perform computationally intensive operations offline while maintaining the capability for rapid online iterative re-planning by only needing to generate and evaluate new candidates when feedback is received.
Data Source
AI summary
Systems and methods for online iterative re-planning are provided herein. In one embodiment, a method includes receiving, at a first time step, a first grasp and an initial object pose of an agent. The method also includes generating a first set of candidate object trajectories based on the first grasp and the initial object pose. Candidate object trajectories of the first set of candidate object trajectories provide a number object poses from the initial object pose to a goal for a number of future time steps after the first time step. The method further includes calculating contact points for grasps associated with each candidate object trajectory of the first set of candidate object trajectories. The method further includes selecting a first candidate object trajectory from the first set of candidate object trajectories. The method includes causing the agent to execute the first candidate object trajectory at a second time step.