HAMRCASTER MOBILITY RESEARCH
A study in mobile robot support mechanics

Caster mobility,
beyond flat ground.

Ball caster × wheelchair casterHardware · perception · control
Simulation model COMPA / caster comparison
Ready to drive Fixed platform yaw
Simplified simulation robot with a circular sensor platform, two rear drive wheels and two front ball casters.

Interactive model requires JavaScript and WebGL. Use the component buttons below to explore the setup.

03
Ball caster support

Two split spherical supports on the front rocker rails.

Drag to rotate · Scroll to zoomIllustrative motion · Face marks heading
The motivation

Small contact.
A different way to move.

How much can a passive support change a robot’s motion? We study ball casters and wheelchair casters on the same mobile platform.

01

Beyond smooth floors

Explore how a rolling spherical support negotiates uneven ground.

02

Resilience to disturbance

Compare resistance to redirection and recovery after a lateral push.

03

More faithful trajectories

Measure how caster motion affects path error and heading stability.

Explore the comparison ↗
One platform. Two support mechanisms.ROS 2 JazzyGazebo simulationEncoder + vision measurement
01 / Hardware overviewSENSE → ESTIMATE → ACT

From a wheel turn to a waypoint.

Encoders, inertial sensing, and depth perception connect the mechanics to an onboard navigation and control loop.

Swipe to explore the diagram →

Hardware sensing and closed-loop control Illustrated encoder, IMU, camera, ESP32, and Raspberry Pi blocks show the hardware. Wheel and turret encoders plus the BNO055 IMU send data through one ESP32 and UART to a Raspberry Pi. A depth camera connects directly by USB. The Pi estimates state and follows waypoint references, sending motor commands to the ESP and wheel and turret drivers. Passive caster encoder boards send independent UDP measurement data to the Pi. Wheel + turret encodersBNO055 IMU RGB-D depth cameraCar ESP32 Raspberry PiDrive + turret Rotation, speed, relative yawHeading and inertial motion Color + registered depthAcquire + timestamp Wheel-speed PIDUART USB · RGB-DMotor drivers + actuation Wheel velocityTurret heading ESP → motor drivers · PWMRate targets → ESP 01 Fuse wheel + IMU state02 Localize with RGB-D 03 Plan / follow waypoints04 Close pose + heading loop Target onboard architecture Independent measurement: H1 / H2 / CENTER encoder boards → Wi-Fi UDP → Pi / experiment logger

Control is layered: the Pi computes trajectory and heading corrections; the car ESP closes the fast wheel-speed PID loop. Visual localization corrects the global reference.

01 / DRIVE

Wheel & turret encoders

The car ESP counts rotation and sends timestamped telemetry over UART.

02 / SENSING

One inertial reference

The same ESP reads the BNO055. A D455 depth camera connects directly to the Pi.

03 / SUPPORT

Interchangeable casters

Split ball casters and conventional swivel casters share the simulation platform.

04 / MEASUREMENT

Observe passive motion

H1, H2, and CENTER encoders record the ball assembly; vision provides another estimate.

See encoder wiring and control loops ↗
02 / Inside the system

Follow the signal.

ROS 2 ↗Nodes, topics, and data flow.VSLAM ↗From depth to a navigable map.Controllers ↗Motion control and encoder sensing.Experiments ↗Matched simulation and field trials.
Publication

A platform for comparison.

Paper details, author links, and final datasets can be added here.

Paper
Title to be added
Authors
Author names · Institution
Venue
Publication details to be added
Resources
Paper · Code · Dataset — links forthcoming