Adaptive Interaction Network for Human Motion Prediction During Human-Robot Collaboration

Area of Science:

• Robotics
• Artificial Intelligence
• Human-Computer Interaction

Background:

• Human motion prediction is vital for safe human-robot collaboration.
• Existing methods often overlook the robot's influence on human movement.
• Robot-aware human motion prediction presents challenges in modeling human-robot heterogeneity and spatial interactions.

Purpose of the Study:

• To develop a novel model for robot-aware human motion prediction.
• To explicitly account for the robot's impact on human trajectories.
• To enhance safety and efficiency in human-robot collaborative environments.

Main Methods:

• An adaptive interaction network (AINet) with two jointly optimized branches: one for human motion and one for robot trajectories.
• A Local-Global Spatial Interaction (LGSI) Module to capture dependencies between human and robot motion sequences.
• An Adaptive Weighted Aggregation (AWA) Module for dynamic feature fusion and a coarse-to-fine prediction strategy.

Main Results:

• The proposed AINet model demonstrates superior performance in human motion prediction tasks.
• Experiments on three datasets validate the effectiveness of the LGSI and AWA modules.
• The method successfully captures fine-grained and global contextual dependencies in human-robot interactions.

Conclusions:

• The developed AINet model significantly advances robot-aware human motion prediction.
• The approach enhances adaptability and accuracy in diverse human-robot collaboration scenarios.
• This work contributes to safer and more efficient human-robot interactions.