Human Movement Prediction for Collision-Aware Robot Workspaces
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current systems for predicting human arm movements in shared workspaces with robots are inefficient due to their reliance on learning-model-based approaches that require extensive training data and struggle in dynamic environments, leading to inaccurate predictions and potential safety hazards.
Innovation Solution
A hybrid data-driven and model-based prediction system that uses a planning algorithm as a generative model, incorporating historical movement information and real-time observations to predict arm movements, considering environmental context, obstacles, and collaborator goals, allowing for real-time predictions without extensive training data.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If learning-model-based approaches are used to predict human arm movements, then the system can handle complex dynamic environments, but the system requires extensive training data and has high computational requirements
Solution Approach 1:
The prediction system is divided into two independent modules: a data-driven component that identifies movement goals and a model-based component (RRT* algorithm) that generates trajectories. This segmentation allows each component to use its strengths without requiring extensive training data, as the model-based planner can generate realistic trajectories from scratch using environmental constraints and robot kinematics.
Solution Approach 2:
The patent introduces an intermediary evaluation mechanism that assesses generated trajectories against observed human movement patterns. This intermediary component filters and selects trajectories that match human behavior without requiring the system to learn from extensive training data, bridging the gap between model-based generation and data-driven accuracy.
2Adaptability or versatility
If learning-model-based approaches are used to predict human arm movements, then the system can adapt to complex environments, but the computational processing time increases
Solution Approach 1:
The RRT* algorithm performs preliminary planning by pre-computing valid trajectories based on environmental constraints and robot kinematics before human movement occurs. This preliminary action creates a library of plausible trajectories that can be quickly evaluated and selected, reducing real-time computational burden while maintaining adaptability to complex environments.
Solution Approach 2:
The system dynamically adjusts the prediction horizon and trajectory evaluation based on the collaborator's current state and environmental context. By making the prediction window adaptive rather than fixed, the system reduces unnecessary computational overhead for distant time points where predictions are less certain, thereby reducing processing time while maintaining accuracy for near-term predictions.
3Measurement precision
If the prediction system considers multiple factors (environmental context, obstacles, collaborator goals), then the prediction accuracy improves, but the device complexity increases
Solution Approach 1:
The patent merges environmental constraint checking, robot kinematics validation, and trajectory optimization into a unified RRT* planning framework. By combining these functions into a single model-based planner, the system achieves high prediction accuracy through multiple factors without proportionally increasing device complexity, as the components work together within one algorithmic structure rather than as separate systems.
Data Source
AI summary
Disclosed herein are systems, devices, and methods for real-time determinations of likelihoods for possible trajectories of a collaborator in a workspace with a robot. The system determines a current kinematic state of the collaborator and determines a goal of the collaborator based on occupancy information about objects in the workspace. The system also determines a possible trajectory for the collaborator based on the goal and the current kinematic state and determines a short-horizon trajectory for the collaborator based on previously observed kinematic states of the collaborator towards the goal. The system also determines a likelihood that the collaborator will follow the possible trajectory based on the short-horizon trajectory, the goal, and the current kinematic state. The system also generates a movement instruction to control movement of the robot based on the likelihood that the collaborator will follow the possible trajectory.