Robot Action Planning Using Predicted Human and Object States
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing robot control systems struggle to account for varying human expectations across different tasks, leading to high design costs and limited versatility, as they often rely on predefined databases or distance-based expressions that are task-specific and difficult to adapt to new scenarios.
Innovation Solution
A robot system equipped with a control device and observation device, utilizing a computer to detect and predict the future states of objects and people, generate action plans based on these predictions, and determine optimal actions considering both human and robot expectations, using a unified model for various tasks.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If a database or computation expression is designed in advance for each task, then the robot can determine appropriate actions for that specific task, but the design cost increases and versatility decreases
Solution Approach 1:
The patent applies universality by creating a unified expectation representation that can be used across multiple tasks. Instead of designing separate databases or computation expressions for each task, the system uses a single expectation representation framework that adapts to different tasks through learning from observation data, allowing the robot to determine appropriate actions for navigation, communication, and other tasks without task-specific predefined models
Solution Approach 2:
The patent applies preliminary action by pre-defining the structure of the expectation representation framework and the learning mechanism, but leaving the specific expectation contents to be learned from data. This allows the system to be prepared for various tasks in advance through a unified framework while adapting to specific task requirements through continuous learning from observation data
2Reliability
If task-specific expectation definitions are created, then the robot can perform that specific task appropriately, but the design cost and complexity increase
Solution Approach 1:
The patent reduces complexity by using a universal expectation representation framework that handles multiple tasks through a single unified structure. This framework learns task-specific expectations from observation data rather than requiring separate predefined definitions for each task, significantly reducing the complexity of expectation definitions while maintaining task-specific accuracy
Solution Approach 2:
The system applies self-service by enabling the robot to automatically learn and adapt expectations for different tasks through the learning mechanism. Instead of requiring manual design of task-specific expectations, the system autonomously learns from observation data, reducing design complexity and enabling automatic adaptation to new tasks
3Productivity
If predefined databases are used for action determination, then the robot can quickly determine actions, but the system lacks adaptability to new scenarios
Solution Approach 1:
The patent applies dynamics by transitioning from static predefined databases to a dynamic learning-based expectation representation. The system continuously learns and updates expectations from observation data, enabling fast action determination through learned patterns while adapting to new scenarios through ongoing learning, combining the speed of predefined systems with the adaptability of learning systems
Data Source
AI summary
A robot system includes a robot controlled by a computer. The robot includes a control device and an observation device. The computer detects a plurality of objects in a periphery by using observation information obtained by observation of the observation device, stores first state information of each of the plurality of detected objects, predicts future states of the plurality of detected objects from the first state information by using a first model for predicting a future state of an object, generates second state information of the periphery obtained by observation of the plurality of detected objects, predicts a future state of the robot from the second state information by using the first model, and determines a future action of the robot based on a given action target, the predicted future states of the plurality of objects, and the predicted future state of the robot.