Hierarchical Robot Command Prediction for Stable Operation Control
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing robot operation systems using trained models struggle to predict stable operations when conditions or environments change, requiring extensive training data and calculation time.
Innovation Solution
An information processing device employs a hierarchical approach with two trained models, where a first model estimates command information over a longer time frame and a second model estimates over a shorter time frame, reducing correlation between input data pieces to stabilize predictions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If a single trained model is used to predict robot operations, then the system is simple, but the prediction becomes unstable when operation conditions or environment change
Solution Approach 1:
The single prediction model is segmented into multiple specialized sub-models, each trained to predict operations under specific operation conditions or environmental parameters. This segmentation allows each sub-model to specialize in particular scenarios, improving overall prediction stability when conditions change, while the modular structure manages complexity through organization rather than monolithic design.
Solution Approach 2:
The system dynamically selects or weights different sub-models based on current operation conditions and environment parameters. This dynamic adaptation allows the system to maintain stable predictions across varying conditions by activating the most appropriate sub-model for the current scenario, transforming a static single-model approach into a dynamic multi-model system.
2Reliability
If more training data is collected to improve prediction stability under changing conditions, then prediction reliability improves, but data collection time and processing complexity increase
Solution Approach 1:
Training data collection is segmented into condition-specific datasets for each sub-model rather than collecting all possible variations for a single model. This allows parallel collection of specialized data for each operation condition or environmental parameter, reducing total collection time while improving prediction stability for each specific scenario through targeted data gathering.
3Productivity
If a single trained model is used, then the system structure is simple, but calculation time increases when handling complex operation variations
Solution Approach 1:
The calculation workload is segmented across multiple specialized sub-models, each handling specific operation conditions or environmental parameters. This distribution of computational tasks reduces the complexity of individual model calculations and improves overall calculation speed by parallelizing processing across condition-specific sub-models rather than using a single monolithic model for all scenarios.
Data Source
AI summary
A first estimation unit (331) estimates command information per first unit of time using a first trained model (321). A second estimation unit (332) estimates, using a second trained model (322), the command information per second unit of time shorter than the first unit of time, from information corresponding to the command information derived by the first trained model (321). An operation control unit (333) operates a control target device (10B) using the command information estimated by the second estimation unit (332).