ModLab · University of Pennsylvania

HAMR

Holonomic mobile robotics

Connecting caster design, visual perception, and feedback control for a holonomic mobile robot. The work spans camera-based caster motion estimation, RGB-D mapping and localization, and motor control.

Camera-based feature tracking used to study the motion of a ball caster
Dual-camera reconstruction of a split spherical caster: shared roll and independent shell rotation.

Caster vision

Track visual features from two cameras to estimate caster motion. The model separates shared rolling motion from independent shell rotation, connecting image measurements with the caster’s mechanics.

Visual SLAM and localization

Use a RealSense D455 with RTAB-Map for RGB-D mapping and localization against saved maps. Wheel odometry and IMU measurements feed an extended Kalman filter to support robot state estimation.

PID control

A Jacobian-based ROS 2 controller translates robot motion commands into drive-wheel and turret commands. ESP32 firmware combines feed-forward terms with PID feedback, with motor response measurements supporting controller tuning.

Ball-caster design

Investigate a split spherical support caster with shared roll and independently rotating hemispheres. Compare its motion with conventional swivel casters through repeatable trajectories and measurements of alignment, slip, and tracking error. These experiments are ongoing.

The latest integrated localization and control setup awaits hardware validation.

Recorded motor response used to inspect and tune the HAMR motor controller

Motor response

Recorded response for inspecting and tuning motor control.

Visualization coming soon

Depth-camera mapping

Map and localization visualization — coming soon.