Robotic Bin Picking with Few-Shot Failure Learning and Perturbation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Robotic devices face challenges in picking items from bins due to physical constraints, varied item positions, materials, and shapes, leading to frequent human intervention and failure in completing tasks.
Innovation Solution
A robotic device equipped with a picking element and a perturbation element, such as a suction mechanism and a compressed air valve, which attempts to pick items, generalizes information from failed attempts, and perturbs the item arrangement to increase the likelihood of successful pickup, using computer vision and machine learning to adapt strategies.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If traditional robotic picking methods are used, then the robotic device can perform picking tasks, but it encounters frequent failures and requires human intervention due to physical constraints and varied item positions
Solution Approach 1:
The robotic device performs self-training by learning from its own failed picking attempts. The system automatically generalizes failure information and transfers it to similar scenarios without requiring human reprogramming or intervention, enabling continuous autonomous improvement of its picking capabilities
Solution Approach 2:
The system implements a feedback loop where failed picking attempts are analyzed, generalized into failure patterns, and used to update the neural network model. This closed-loop learning process continuously improves task completion rates by incorporating real-world failure data back into the training process
2Reliability
If extensive training data is used to train the artificial neural network, then the model achieves better performance, but the training process becomes extremely time-consuming
Solution Approach 1:
The system performs preliminary generalization of failure information during the picking process itself. By extracting and generalizing failure patterns in real-time and transferring them to similar scenarios, the system prepares learning insights proactively rather than requiring extensive post-training data collection and processing
Solution Approach 2:
The system creates generalized representations of failure scenarios that can be copied and applied to similar picking situations. Instead of requiring unique training data for every possible failure mode, the system generalizes from specific failures to create reusable knowledge patterns that transfer across multiple scenarios
3Reliability
If the robotic device attempts to pick items from all possible positions, then it may find successful pickup points, but the time and attempts required increase significantly
Solution Approach 1:
The system performs preliminary identification of potential failure regions using the trained neural network before actual picking attempts. By pre-assessing which regions are likely to fail based on generalized failure patterns, the system avoids unnecessary picking attempts and directs efforts toward more promising locations
Solution Approach 2:
Instead of searching for successful pickup points directly, the system inverts the approach by first identifying and avoiding failure regions. The neural network is trained to recognize failure patterns, and the robotic device strategically selects picking locations that minimize the probability of failure rather than exhaustively testing all positions
Applied Scientific Principles
This section explains which scientific principles are used to turn an abstract innovation direction into a practical engineering solution.
Function Achieved in This Case
The robotic device can effectively pick items from bins with varying arrangements by perturbing items to reposition them for easier access, reducing the need for human intervention and improving task completion rates.
Implementation Method 1
The robotic device includes a picking element and a perturbation element
Implementation Method 2
The perturbation element may be used to blow high-pressure air into the bin in order to perturb the contents of the bin
Data Source
AI summary
Various embodiments of the present technology generally relate to robotic devices and artificial intelligence. More specifically, some embodiments relate to an artificial neural network training method that does not require extensive training data or time expenditure. The few-shot training model disclosed herein includes attempting to pick up items and, in response to a failed pick up attempt, transferring and generalizing information to similar regions to improve probability of success in future attempts. In some implementations, the training method is used to robotic device for picking items from a bin and perturbing items in a bin. When no picking strategies with high probability of success exist, the robotic device may perturb the contents of the bin to create new available pick-up points. In some implementations, the device may include one or more Computer-vision systems.