Physics-Based Control Parameter Planning for Autonomous Systems
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional control methods for autonomous technical systems, such as robots, struggle to predict the physical effects of their actions on the environment, limiting their ability to determine optimal action sequences due to a lack of consideration for physical laws and interactions.
Innovation Solution
A method that uses a physical simulation model and machine learning to simulate various action steps, training a routine to evaluate and optimize control parameters, allowing for flexible and efficient determination of favorable action sequences by accounting for physical effects like gravity and collisions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Loss of information
If conventional control methods based on sensor detection and path planning algorithms are used, then the system can determine action sequences, but the system cannot predict physical effects and interactions with the environment
Solution Approach 1:
A physics simulation environment is introduced as an intermediary between the control system and the real environment. This virtual environment predicts physical effects and interactions before actual actions are executed, allowing the system to learn optimal control strategies without directly experiencing physical consequences.
Solution Approach 2:
A virtual copy of the physical environment is created through physics simulation. This digital twin replicates physical laws, object properties, and environmental conditions, enabling the system to train and test control strategies in a realistic yet controllable virtual space before deploying to the real system.
2Reliability
If physics-based simulation and machine learning training are used to determine optimized action sequences, then the system can account for physical effects, but the computation time and processing requirements increase
Solution Approach 1:
The system performs preliminary training in the physics simulation environment before actual deployment. By pre-training the machine learning model with extensive simulation data that incorporates physical laws, the system accumulates knowledge about physical effects in advance, so that during real operation, decisions can be made quickly without extensive real-time computation.
Solution Approach 2:
The system uses itself to generate training data through autonomous exploration in the physics simulation environment. The technical system autonomously performs actions, observes physical consequences, and uses this self-generated data to train its own control policies, eliminating the need for external manual data collection and labeling.
3Adaptability or versatility
If the system explores various combinations of action steps through simulation, then optimized control strategies can be found, but the number of simulations required increases computational complexity
Solution Approach 1:
The system dynamically adjusts the complexity and scope of simulations based on the current state and learning progress. Instead of exhaustively exploring all possible action sequences from the beginning, the simulation complexity evolves as the system learns, allowing flexible exploration of action spaces while managing computational resources efficiently.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
The invention relates to a method and a device for ascertaining control parameters in a computer-assisted manner for advantageously handling a technical system. A starting state (AZ) and the surroundings (UM) of the technical system (TS) are detected using at least one sensor (101), and a physical simulation model (SIM) of the technical system is generated using same. On the basis of the starting state (AZ), different combinations of handling steps of the technical system (TS) are simulated with respect to a specified target state (ZZ) using the simulation model (SIM), wherein control parameters of the technical system for carrying out the handling steps are varied. The simulation data is used to train a machine learning routine by means of an evaluation of each handling step, and the trained machine learning routine is used to ascertain an optimized combination of handling steps. The control parameters of the optimized combination of handling steps are output in order to control the technical system (TS).