Introduction
In physical Human-Robot Interaction (pHRI), robots must balance compliance with human guidance against strict adherence to task-specific constraints, such as safety boundaries or assembly paths. While Hierarchical Quadratic Programming (HQP) is a standard tool for managing these competing priorities, it often suffers from a "direction problem." When a robot encounters a constraint, HQP may satisfy the mathematical requirements by moving the robot in an arbitrary direction, leading to unpredictable behavior that the human operator must fight.
A recent study proposes a novel control framework that introduces "directional constraints" to hierarchical tasks. This approach ensures that the robot's motion remains aligned with human intention even when navigating complex constraints, facilitating smoother and more intuitive collaboration.
The Challenge of Hierarchical Tasks
In collaborative scenarios, a robot typically manages a hierarchy of tasks. For instance, a high-priority safety task (e.g., collision avoidance) might override a lower-priority interaction task (e.g., following human force). Standard HQP solvers satisfy high-priority constraints by allowing the robot's velocity to point in any feasible direction. This often results in "slipping" or jerky movements.
Existing "task scaling" methods attempt to maintain the human's intended direction exactly. However, if the intended path is blocked, the robot stops entirely, causing frustration. The authors identified the need for a middle ground: a system that allows slight directional adjustments to bypass constraints without losing the intuitive "feel" of human guidance.
Proposed Methodology
The researchers developed a three-part framework to address these issues:
- Directional Constraints: Instead of requiring exact directional alignment, the framework defines a "cone" of acceptable motion. This angular range allows the robot the flexibility to satisfy high-priority constraints while staying generally aligned with the human's desired path.
- Direction-Constrained Optimization Algorithm: To handle the non-linear nature of directional constraints, the authors designed an algorithm that splits the optimization into two parallel sub-problems: a "Minimum Error Search" (standard HQP) and a "Minimum Angle Search" (minimizing deviation from the intended direction). The solutions are then merged to find an optimal, feasible velocity.
- Variable Admittance Control: Standard admittance controllers can suffer from "wind-up" when a human pushes against a constraint, leading to lag when the human tries to pull away. The proposed framework introduces variable damping that increases when the robot is restricted by a constraint, preventing this lag and improving responsiveness.
Experimental Evaluation
The framework was tested using a 7-degree-of-freedom UFACTORY xArm7 robotic arm in several scenarios:
- Velocity-constrained tasks: Navigating within speed limits.
- Position-constrained tasks: Moving around virtual walls.
- Human-Robot Co-Assembly: A real-world task involving the alignment of aluminum profiles.
Key Results and Insights
The experimental results demonstrated significant improvements over traditional HQP and task scaling methods:
- Trajectory Smoothness: The proposed method eliminated oscillations and jitter commonly seen in standard HQP during constrained motion.
- Boundary Efficiency: In virtual wall experiments, while task scaling caused the robot to be blocked for approximately 5 seconds, the proposed method allowed for smooth movement along the wall with near-zero blocking time.
- Reduced Interaction Delay: The variable admittance control reduced the interaction delay during direction changes from 1.0 second to 0.55 seconds.
- Practical Utility: The system successfully assisted in complex assembly, maintaining critical orientations while allowing the human to guide lateral positioning intuitively.
Conclusion
By relaxing strict directional requirements into a flexible angular range, the proposed framework allows robots to "slide" along constraint boundaries smoothly. This approach, combined with variable damping, significantly enhances the efficiency and feel of pHRI. Future research will focus on automating the tuning of parameters like the "cone" angle and exploring performance in highly cluttered environments with overlapping constraints.
Summary updated