Collaborative Robot Training Using Shared Demonstration Trajectories
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional methods for training multiple robots to perform collaborative tasks are inefficient and cumbersome, requiring separate training for each robot and extensive expertise in machine learning and simulation environment creation, which limits scalability and accessibility.
Innovation Solution
A methodology that uses a single machine learning model to train multiple robots simultaneously through visual analytics and image learning algorithms, collecting expert demonstrations with cameras and markers to acquire reference trajectories and generate a simulation environment, allowing for reduced user demonstrations and simplified training of collaborative tasks.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If traditional robotic programming methods are used to train multiple robots, then each robot can be programmed to perform its specific task, but the complexity and time required increases significantly as the number of robots increases
Solution Approach 1:
The patent merges the training process of multiple robots into a single unified demonstration. Instead of programming each robot separately, a single expert demonstration captures the collaborative task, and the system automatically generates coordinated trajectories for all robots involved, reducing programming complexity while maintaining task reliability
Solution Approach 2:
The patent introduces an intermediary system that processes expert demonstrations and generates robot trajectories. This intermediary layer (the training system with simulation environment and trajectory generation algorithms) mediates between the expert demonstration and the actual robot execution, simplifying the programming process while ensuring reliable task execution
2Manufacturing precision
If separate training is performed for each robot, then each robot learns its specific role, but the number of expert demonstrations required increases exponentially
Solution Approach 1:
The patent combines the training data collection for multiple robots into a single expert demonstration. The system captures trajectories for all robots simultaneously during one demonstration, then processes this data to generate coordinated training sets for each robot, dramatically reducing the time and number of demonstrations needed while maintaining precise task execution
Solution Approach 2:
The patent performs preliminary processing of expert demonstrations to extract and organize trajectories for multiple robots in advance. By pre-processing the demonstration data to identify and separate individual robot trajectories from the collaborative task, the system eliminates the need for multiple separate demonstrations while ensuring precise task execution
3Adaptability or versatility
If conventional training methodologies are used, then robots can be trained for collaborative tasks, but extensive expertise in machine learning and simulation environment creation is required
Solution Approach 1:
The patent enables the system to automatically create simulation environments and process training data without requiring expert intervention. The system self-configures the simulation environment based on the expert demonstration and automatically generates training datasets, making the training process accessible to non-specialized personnel while maintaining collaborative task capability
Solution Approach 2:
The patent introduces an intermediary training system that handles the complexity of machine learning and simulation environment creation. This intermediary layer processes expert demonstrations and automatically generates ready-to-use training data and environments, shielding users from technical complexity while enabling versatile collaborative task training
4Reliability
If multiple expert demonstrators are used to train each robot separately, then comprehensive task coverage is achieved, but the process becomes cumbersome and inefficient
Solution Approach 1:
The patent merges the functionality of multiple expert demonstrators into a single demonstration. By capturing trajectories for all robots in one collaborative demonstration and then automatically processing the data to extract individual trajectories, the system achieves comprehensive task coverage with a single demonstration, dramatically improving training efficiency
Data Source
Figure 1~2
Figure 3
Figure 4
AI summary
The present disclosure provides describes to train a multi policy ML model to control robots in a multi-robot system in collaborating to perform a task. For example, trajectories associated with manipulating an object to perform the collaborative task can be determined and an ML model trained to output control actions for the robots in the multi-robot system to collaborate to complete the task.