JAMES WU
DESIGN ENGINEER
21 / Physical / Course project

Applied Robotics · Planning and Interaction

Third-year robotics labs using DE NIRO and Baxter in simulation. I worked through mobile-robot planning, dual-arm handover, Jacobian-based velocity control, redundancy resolution and contact tasks.

PROJECTApplied Robotics · Planning and Interaction
INDEPENDENT WORKIndependent lab implementation and analysis
CONTEXTIndependent labs · Lab 2 and Lab 3
2026 · Year 3
Planning and control workflow: configuration space, roadmap, robot interactionInspect full-size image ↗
FIG. 01A reading map of the two documented simulation labs.
01 / OVERVIEW

The project in a minute.

THE OUTCOME

Two connected lab studies: compare planners on an indoor map, then use position, velocity and torque control for manipulation and interaction.

What supports this? ↓
The challenge
How do map representation, planning and control choices change what a robot can actually execute?
Independent work
Independent lab implementation and analysis. The planning report compares 22.22 m manual waypoints with an 18.58 m PRM route. The interaction report documents null-space obstacle avoidance and a wiping-controller fix using true angular velocity.
Read this in context
Individually completed course labs, confirmed by James. Results are simulation-specific; Dijkstra is optimal on the sampled graph, not guaranteed globally optimal in continuous space.
01Map + footprintCreate collision-aware C-space.02Plan + compareWaypoints, potential field and PRM.03Control + interactPosition, velocity and torque tasks.
02 / THINKING

The choices behind the result.

Open a chapter to follow the work and the choices made along the way.

01 / DESIGN

Plan for a body, not a point

Inflate obstacles using the robot footprint, then compare hand-picked waypoints, potential fields and PRM plus Dijkstra on the same indoor map.

Inspect the supporting evidence ↓
02 / ENGINEERING

Judge a route by more than its length

The reported geometric route changes from 22.22 m to 18.58 m. Graph density and connections affect this result; sharper turns and lower clearance can still make tracking harder.

Inspect the supporting evidence ↓
03 / ENGINEERING

Use the redundant joints deliberately

Jacobian-based velocity commands and null-space projection combine an end-effector task with an elbow obstacle-avoidance objective.

Inspect the supporting evidence ↓
04 / ENGINEERING

Damp rotation using the correct feedback

For sponge wiping, the corrected controller takes angular velocity from J(q)·q̇ rather than orientation values. Adding rotational damping improved the simulated contact posture.

Inspect the supporting evidence ↓
METHODS USEDROS / GazeboC-spacePRM / DijkstraJacobian controlNull-space projectionTorque control
03 / EVIDENCE

Results, and what they mean.

Individually completed course labs, confirmed by James. Results are simulation-specific; Dijkstra is optimal on the sampled graph, not guaranteed globally optimal in continuous space.

22.22 → 18.58 m

Reported geometric route length

Manual waypoints → PRM + Dijkstra on the same map; approximately 16.4% shorter, not a travel-time gain.

3

Control modes

Position for transfer, velocity for path following, torque for dynamic/contact tasks.

Simulation

Validation setting

DE NIRO / Baxter course tasks completed individually in simulation.

Source notes & scope 3
  1. Robotics Lab 2pp. 4–29

    C-space, planner implementation and route-length comparison on pp. 25–26.

  2. Robotics Lab 3 — Interactpp. 3–36

    DE NIRO / Baxter manipulation, redundancy and contact control; wiping correction on pp. 34–35.

  3. Author clarificationLab ownership

    Labs 1, 2 and 3 were each completed individually; shared report cover names do not define implementation ownership.

↑ Return to the overview