Robotic Excavation Planning for Rigid Objects in Clutter
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Autonomous excavation for hard or compact materials, especially irregular rigid objects, remains challenging due to high variance in geometrical shapes and physics properties, leading to increased difficulty in robotic perception and manipulation, and existing technologies have not effectively addressed this issue in cluttered environments.
Innovation Solution
The development of novel RGBD and voxel-based convolutional neural network (CNN) models for predicting excavation success and formulating excavation planning as an optimization problem, using a 3D voxel-grid representation of the excavation scene to overcome the simulation-to-real gap and improve robotic excavation efficiency.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If traditional excavation methods are used for rigid objects in clutter, then the system can operate with simple algorithms, but the excavation success rate is low and time-consuming
Solution Approach 1:
The system performs preliminary actions by predicting excavation outcomes before actual excavation occurs. The outcome prediction module forecasts whether an excavation will succeed or fail, allowing the system to plan ahead and avoid unsuccessful excavations, thereby improving success rates while reducing wasted time.
Solution Approach 2:
The system dynamically adapts its excavation planning based on real-time predictions. The optimization module uses predicted outcomes to dynamically adjust the excavation sequence, selecting the most promising targets first and revising plans as conditions change, enabling efficient time utilization while maintaining high success rates.
2Manufacturing precision
If complex prediction models are used to improve excavation planning, then the excavation quality improves, but the computational complexity increases
Solution Approach 1:
The complex prediction task is segmented into multiple independent prediction modules, each handling specific aspects of excavation outcomes. This segmentation allows the system to achieve high prediction accuracy through specialized sub-models while managing computational complexity by dividing the overall problem into smaller, more tractable components.
Solution Approach 2:
The system introduces an intermediary optimization module that bridges the prediction models and final excavation execution. This intermediary translates complex prediction outputs into optimized excavation sequences, enabling high-quality trajectories without directly increasing the complexity of individual prediction models.
3Quantity of substance
If more excavation attempts are made to improve success rate, then the excavation completeness improves, but the resistive force and time consumption increase
Solution Approach 1:
The system performs preliminary outcome predictions to identify which excavation attempts are likely to succeed before executing them. By forecasting results in advance, the system can prioritize excavations with high predicted success rates, achieving greater total excavation volume while avoiding unnecessary attempts that would increase resistive forces and time consumption.
Solution Approach 2:
The optimization module selectively performs only the necessary excavation attempts required to achieve high success rates, avoiding excessive actions. By using prediction-guided selection, the system achieves sufficient excavation completeness with fewer attempts, thereby reducing cumulative resistive forces and time consumption.
Data Source
AI summary
Embodiments of a learning-based excavation planning method are disclosed for excavating rigid objects in clutter, which is challenging due to high variance of geometric and physical properties of objects, and large resistive force during the excavation. A convolutional neural network is utilized to predict a probability of excavation success. Embodiments of a sampling-based optimization method are disclosed for planning high-quality excavation trajectories by leveraging the learned prediction model. To reduce simulation-to-real gap for excavation learning, voxel-based representations of an excavation scene are used. Excavation experiments were performed in both simulation and real world to evaluate the learning-based excavation planners. Experimental results show that embodiments of the disclosed method may plan high-quality excavations for rigid objects in clutter and outperform baseline methods by large margins.