Robot Pickup Control Using Depth Images and Success Probabilities
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Robotic devices face challenges in efficiently picking up and moving objects due to dynamic and geometric conditions, such as other objects and geometric configurations, which can obstruct their trajectory and limit their functionality.
Innovation Solution
A method and device that utilize a trained control model to determine optimal robot configurations and trajectories by considering dynamic conditions, including relationships between objects and geometric conditions, using a combination of robot trajectory, precondition, and end condition models learned from demonstrations, and reinforced training with depth images to improve the accuracy of pickup and movement success probabilities.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If a robot control model ascertains control commands considering other objects and geometric conditions, then the reliability of object pickup and movement is improved, but the computing complexity and time required for trajectory calculation increase
Solution Approach 1:
The system performs preliminary training of the robot trajectory model and precondition model using depth images and demonstrated pickup actions before actual operation. This pre-computed knowledge is stored and reused during runtime, allowing the robot to quickly determine control commands without performing complex trajectory calculations in real-time, thus improving reliability while reducing computation time during actual pickup tasks
Solution Approach 2:
The system uses depth images to create a digital representation of the environment and objects, and uses demonstrated pickup actions as training data to create a model of successful pickup behaviors. These copied representations and patterns are then used by the trained models to determine control commands, avoiding the need to recalculate from first principles each time a pickup is performed
2Manufacturing precision
If the robot control model is adapted to geometric and dynamic conditions through training, then the manufacturing precision of pickup operations is improved, but the training time and computational resources required increase
Solution Approach 1:
The robot trajectory model and precondition model are trained in advance using collected depth images and demonstrated pickup actions before the robot begins actual pickup operations. This preliminary training phase allows the models to learn optimal pickup configurations for various objects and conditions, improving precision without adding computation time during actual pickup tasks
Solution Approach 2:
The system uses the robot's own demonstrated pickup actions as training data, allowing the model to learn from its own experiences. The robot performs pickup actions, collects depth images and action data, and uses this self-generated data to train and improve its control models, reducing the need for external training resources
3Reliability
If multiple pickup robot configurations are evaluated with probability of success, then the reliability of object pickup is improved, but the device complexity of the control model increases
Solution Approach 1:
The system uses depth images to create a digital copy of the environment and objects, and uses demonstrated pickup actions as training data. The trained robot trajectory model processes these copied representations to evaluate multiple pickup configurations and their probabilities of success, simplifying the control model structure while maintaining reliability
Solution Approach 2:
The system evaluates multiple pickup robot configurations and selects those with high probability of success based on the trained model's assessment. The model uses feedback from the depth images and previously demonstrated actions to determine which configurations are most likely to succeed, allowing reliable pickup without requiring overly complex control models
Data Source
AI summary
A device and a method for controlling a robotic device, including a control model. The control model includes a robot trajectory model, which for the pickup includes a hidden semi-Markov model with one or multiple initial states, a precondition model, which for each initial state of the robot trajectory model includes a probability distribution of robot configurations before the pickup is carried out, and an object pickup model, which for a depth image outputs a plurality of pickup robot configurations having a respective associated probability of success.