Overview

This project explores camera calibration, object detection, and robot control on a 5‑DOF RX200 robot arm. The system detects blocks and uses forward and inverse kinematics to pick up, stack, and sort them. It also includes a custom basketball launcher that picks up small basketballs and throws them into baskets. Through this experience, I gained insights into the practical aspects of manipulation laying the foundation for my own future robotics experiences.

Background and goals

Fig. 1. Workspace of our RX200 arm system

The setup includes a 5‑DOF RX200 arm, an Intel RealSense L515 camera, and an Intel i7 for compute. Goals:

• Detect and classify blocks
• Calibrate the camera (intrinsics, extrinsics, and homography)
• Execute pick, place, stack, and swap behaviors
• Build a basketball launcher and tune motion for accurate throws

Team: David Ho, Atharva S. Kashyap, Hun Kuk Park, Christopher Wu

Contact: {homandat, katharva, parkcart, chriswuc}@umich.edu

Methods

Cycle task

A human places blocks at (−100, 225) and (100, 225). The robot swaps them using an intermediate point at (250, 75). Using “teach and repeat,” the arm is torque‑off while a person guides it through waypoints and gripper actions, which are recorded and replayed.

Calibration

1. Intrinsics: RealSense2 + calibration node + checkerboard, sweeping the FOV and tilting to capture skew. Repeated three times until convergence.

2. Extrinsics:

• Manual estimate from measured camera pose relative to the board