Robot Subgoal Setting for Lower-Complexity Operation Planning
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing robot control methods face high computational complexity in determining optimal operation sequences, leading to increased time steps required to complete tasks.
Innovation Solution
A control device and method that utilize a subgoal setting mechanism to abstract workspace states, setting intermediate goals and constraints, followed by an operation sequence generation process to optimize task execution.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional robot control methods are used to determine optimal operation sequences, then task completion accuracy is improved, but computational complexity increases enormously and time steps required increase
Solution Approach 1:
The patent segments the complex operation sequence determination into two distinct phases: (1) generating multiple candidate operation sequences using heuristic methods, and (2) selecting the optimal sequence using reinforcement learning. This segmentation reduces computational complexity by avoiding exhaustive search while maintaining task completion accuracy through the two-stage optimization process.
Solution Approach 2:
The patent performs preliminary action by pre-generating multiple candidate operation sequences using heuristic methods before applying reinforcement learning for optimal selection. This preliminary generation of candidates reduces the computational burden of the subsequent optimization phase, allowing the system to efficiently determine the best sequence without exhaustive search.
2Measurement precision
If traditional robot control methods are used to determine optimal operation sequences, then task completion accuracy is improved, but the number of time steps required to complete the task increases
Solution Approach 1:
The patent segments the operation sequence determination into candidate generation and optimal selection phases, enabling faster computation of viable options through heuristics while reserving computationally intensive optimization for selective refinement, thereby reducing overall time steps required.
Solution Approach 2:
By pre-generating multiple candidate sequences using efficient heuristic methods before applying reinforcement learning, the system avoids exhaustive search in real-time, significantly reducing the number of time steps required to determine the optimal operation sequence while maintaining high task completion accuracy.
3Measurement precision
If exhaustive search methods are used to find optimal operation sequences, then solution optimality is improved, but computational complexity becomes enormous
Solution Approach 1:
The patent divides the exhaustive search problem into two parts: heuristic generation of candidate sequences (reducing search space) and reinforcement learning-based optimization (ensuring optimality). This segmentation achieves solution optimality without the enormous computational complexity of complete exhaustive search.
Solution Approach 2:
The patent introduces reinforcement learning as an intermediary mechanism between candidate generation and final selection. This intermediary learns optimal selection policies from experience, enabling the system to achieve solution optimality without performing exhaustive search, thereby reducing computational complexity significantly.
Data Source
AI summary
A control device 1B mainly includes a subgoal setting means 17B and an operation sequence generation means 18B. The subgoal setting means 17B is configured to set a subgoal “Sg” based on abstract states in which states in a workspace where a robot works are abstracted, the subgoal Sg indicating an intermediate goal for achieving a final goal or constraint conditions required to achieve the final goal. The operation sequence generation means 18B is configured to generate an operation sequence to be executed by the robot based on the subgoal.