Pick-and-Place Challenge

This is a course project for MEAM 5200 Introduction to Robotics, taught by Prof. Rachel Holladay. [report]

In this project, we develop a unified manipulation pipeline capable of autonomously detecting, grasping, and stacking both static and dynamic blocks to maximize the final score under strict safety and interface constraints.

Our approach leverages end-effector–mounted vision to estimate block poses via AprilTag detections and transforms these observations into the robot base frame. For static blocks, the system performs sequential perception-driven pick-and-place actions, aligning end-effector orientation with block geometry and stacking objects incrementally on the goal platform. For dynamic blocks, we implement a predictive grasping strategy that estimates angular velocity from multiple time-stamped observations, forecasts the future block pose, and synchronizes robot motion with the estimated arrival time to enable successful closed-loop interception.