Robot Manipulation Learning From Demonstrations in Dynamic Environments

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Robots lack the ability to perform tasks in unstructured environments without pre-programmed manipulation skills, as they struggle to adapt to dynamic and uncertain conditions, especially when interacting with humans and objects.

Innovation Solution

A robotic system capable of learning skills through human demonstrations and interactions, using machine-learning techniques to acquire and execute manipulation skills in unstructured environments, without requiring pre-programmed manipulation skills, and can adapt to changes in its surroundings.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Extent of automation

If robots are equipped with manipulators to perform tasks autonomously, then task automation capability is improved, but programming complexity and device complexity increase significantly

Engineering Contradiction:
Improvetask automation capabilityVSAvoidprogramming complexity
Core Design Contradiction:
Extent of automationVSDevice complexity

Solution Approach 1:

The robot performs self-learning by observing human demonstrations and automatically generating its own control policies through reinforcement learning, eliminating the need for complex pre-programming. The system serves itself by acquiring manipulation skills autonomously through interaction with the environment and receiving rewards for successful task completion.

Inventive Principle:
Principle #25Self-service

Solution Approach 2:

The system pre-trains policies in simulation environments before deploying them in real-world scenarios. This preliminary training in controlled virtual settings allows the robot to learn basic manipulation skills and adapt to various conditions before facing actual physical tasks, reducing the complexity of direct real-world programming.

Inventive Principle:
Principle #10Preliminary action

2Reliability

If robots are designed to operate in structured environments with pre-programmed skills, then operational reliability is improved, but adaptability to unstructured environments deteriorates

Engineering Contradiction:
Improveoperational reliabilityVSAvoidadaptability to unstructured environments
Core Design Contradiction:
ReliabilityVSAdaptability or versatility

Solution Approach 1:

The robot employs dynamic policy adaptation by continuously learning from human demonstrations and environmental feedback. Instead of relying on fixed pre-programmed skills, the system dynamically adjusts its behavior through reinforcement learning, allowing it to adapt to unstructured environments while maintaining operational reliability through continuous improvement.

Inventive Principle:
Principle #15Dynamics

Solution Approach 2:

The system changes its internal parameters and policies based on observed human behavior and environmental conditions. By modifying its control parameters through learning algorithms, the robot can adapt to different unstructured environments and tasks while maintaining reliable operation through data-driven parameter optimization.

Inventive Principle:
Principle #35Parameter changes

3Adaptability or versatility

If robots continuously acquire information about the environment to make autonomous decisions, then adaptability to dynamic environments is improved, but computational load and processing time increase

Engineering Contradiction:
Improveadaptability to dynamic environmentsVSAvoidprocessing time
Core Design Contradiction:
Adaptability or versatilityVSLoss of time

Solution Approach 1:

The system pre-processes and learns from environmental information during demonstration phases and simulation training, building pre-trained policies that encode environmental understanding. This preliminary information processing allows the robot to make rapid autonomous decisions during execution without real-time computational delays.

Inventive Principle:
Principle #10Preliminary action

Solution Approach 2:

The robot creates simplified representations or copies of complex environmental states through learned models and simulations. By working with these compressed representations rather than raw sensor data, the system reduces computational load while maintaining adaptability to dynamic environments through efficient pattern recognition.

Inventive Principle:
Principle #26Copying

Data Source

PatentEP3585569B1Systems, apparatus, and methods for robotic learning and execution of skills
Publication Date: 2023.06.07 DILIGENT ROBOTICS INC
  • EP3585569B1 patent drawingFigure 1
  • EP3585569B1 patent drawingFigure 2
  • EP3585569B1 patent drawingFigure 3

AI summary

Systems, apparatus, and methods are described for robotic learning and execution of skills. A robotic apparatus can include a memory, a processor, a manipulating element, and sensors. The processor can be operatively coupled to the memory, the manipulating element, and the sensors, and configured to: obtain a representation of an environment; identify a plurality of markers in the representation of the environment that are associated with a set of physical objects located in the environment; obtain sensory information associated with the manipulating element; and generate, based on the sensory information, a model configured to define movements of the manipulating element to execute a physical interaction between the manipulating element and the set of physical objects. Alternatively or additionally, the processor can be configured to generate, using the model, a trajectory for the manipulating element that defines movements of the manipulating element associated with executing the physical interaction.