Model-Predictive End-Effector Selection for High-Throughput Bin Picking
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Automated bin picking of diverse objects is challenging due to the complexity of selecting the appropriate end-effector, which involves high-dimensional visual sensory inputs and uncertainties, leading to increased cycle times and reduced throughput.
Innovation Solution
A computer system uses a model-predictive control algorithm and Markov Decision Process (MDP) to optimize end-effector selection, incorporating pre-trained grasp prediction models and void zones to plan efficient tool changes, employing integer linear programming (ILP) or sparse tree search for real-time decision-making.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If the robot frequently changes end-effectors to match diverse objects, then the adaptability to handle various object sizes and shapes is improved, but the cycle time increases and throughput decreases
Solution Approach 1:
The system performs preliminary planning by predicting future object picks and pre-selecting end-effectors before they are needed. The lookahead planner anticipates the sequence of picks and prepares tool changes in advance, reducing idle time and improving throughput while maintaining adaptability.
Solution Approach 2:
The end-effector selection is made dynamic and adaptive through the lookahead planner that continuously evaluates future pick sequences. The system dynamically adjusts tool changing decisions based on predicted future needs rather than following a fixed schedule, optimizing the balance between adaptability and throughput.
2Reliability
If the robot performs thorough end-effector selection and tool changing, then the pick success rate is improved, but the cycle time increases
Solution Approach 1:
The system performs preliminary end-effector selection and tool changing planning before picks are executed. By predicting future object characteristics and pre-selecting appropriate end-effectors, the system ensures high pick success rates while minimizing the actual time spent on tool changes during execution.
Solution Approach 2:
The system skips unnecessary tool changes by using the lookahead planner to identify sequences where the current end-effector can handle multiple future objects. This reduces the frequency of tool changes while maintaining pick success rates by ensuring the right tool is selected in advance when needed.
3Productivity
If the system uses complex algorithms for end-effector selection, then the optimization of throughput is improved, but the computational complexity and decision-making time increase
Solution Approach 1:
The complex decision-making process is segmented into separate modules: a reward function that evaluates pick success probability, a lookahead planner that predicts future picks, and an execution component. This segmentation allows each module to specialize in specific tasks, improving overall throughput optimization while making the system more manageable and efficient.
Solution Approach 2:
The system skips exhaustive search of all possible end-effector sequences by using the lookahead planner to focus computation on the most promising future picks. This selective approach optimizes throughput by concentrating computational resources on critical decision points rather than evaluating all possibilities.
Data Source
AI summary
A computer system automatically selects robot end-effectors for pick-and-place applications using a model-predictive control algorithm. The system may select an end-effector to replace an existing end-effector in order to optimize (or at least increase) throughput. The system uses a predictive model of reward, where reward of each potential grasp for each end tool is parameterized by a deep neural network. The system may also use a variety of metrics to evaluate the performance of the tool-selection algorithm, and thereby improve performance of the system.