This repository presents a perception-driven orientation control framework for robotic inspection tasks. The system integrates real-time point cloud perception, orientation error estimation, and an admittance-based control strategy to enable compliant and stable end-effector alignment with target surfaces.
The framework is designed for human-in-the-loop robotic inspection, where the controller assists operators by autonomously regulating orientation while respecting physical constraints such as torque and velocity limits.
- Perception-driven control using surface normals from point clouds
- Admittance-based outer-loop controller for compliant behavior
- PD-based orientation regulation with physically interpretable dynamics
- Compatible with ROS 2 and MoveIt Servo pipelines
- Real-time integration of perception, control, and execution
- Validated on UR5e manipulator hardware
The control pipeline consists of the following stages:
Depth Image -> Point Cloud -> Surface Normal Estimation
-> Orientation Error Computation
-> PD Controller (Torque Output)
-> Admittance Dynamics (Virtual Sphere Model)
-> Velocity Commands -> Robot Execution