Master of Engineering · 2026

Dynamic Robot Motion Planning

A comparative evaluation of RRT-based replanning in changing environments, supported by controlled simulation and physical validation.

RRTRRT-ConnectRRT*RRTXFour planners under the same dynamic scenario

System developed

One framework for measuring how planners adapt.

The project placed four sampling-based planners inside a common discrete-time replanning loop. Each received the same start, goal, occupancy map, obstacle updates, and computational budget so differences in speed, path quality, clearance, and replanning burden could be compared fairly.

Motion planningComputer visionSimulationCalibrationData analysis

Project storyline

From planner theory to a live projected path.

Frame the comparison

Evaluate adaptation, not just first-path generation.

The environment changed frame by frame. A previously valid route could become blocked, forcing each planner to continue growing, reset and replan, or repair its existing structure. The study measured both planning effectiveness and the cost of adaptation.

RRTExploratory baseline

RRT-ConnectBidirectional search

RRT*Path refinement

RRTXRepair-oriented replanning

Build a common loop

Give every planner the same representation and budget.

Both the physical and simulated systems reduced their environment to a 2D occupancy mask. The planner received a start state, goal state, map dimensions, and obstacle mask, then expanded its search structure for a fixed budget before the next update.

Observe environment→Build occupancy mask→Update planner→Log performance

Validate physically

Connect perception, planning, and projection.

A top-down camera detected the start marker, goal marker, and warehouse-style obstacles using color-based segmentation. Camera-projector calibration mapped computed paths back onto the physical surface as a live overlay.

Physical workspace with projected paths and warehouse-style obstacles — Projected planning workspace

Overhead camera and projector used for physical validation — Top-down sensing

Camera-projector dot calibration process — Projection calibration

Benchmark in simulation

Control the environment for repeatable comparison.

The benchmark used a 1000 × 700 workspace, three moving obstacles, 260 frames, and 120 planning steps per frame. Twenty random seeds produced averaged metrics while a synchronized four-panel view made the evolving search structures visible.

Dynamic four-planner comparison

Representative simulation frame with moving obstacles and planned path — Common simulation environment

Measure the tradeoffs

Fast solutions and short paths came from different planners.

RRT, RRT-Connect, and RRT* each solved all 20 simulation seeds. RRT-Connect reached feasible solutions fastest. RRT* produced the shortest average paths, but required substantially more planning time and more replanning events.

6,095 msRRT-Connect average first solution

975 pxRRT* average path length

10.19 pxRRT average minimum clearance

Interpret the result

The best planner depends on the objective and environment.

RRT-Connect was strongest for rapid feasible replanning. RRT* was strongest when path refinement justified extra computation. RRT remained a reliable baseline with greater average clearance. RRTX highlighted how strongly repair-based behavior can depend on the way environmental change is represented.

Average time to first solution by planner

4RRT-family planners

20simulation seeds

260dynamic frames

120planning steps per frame

Team

George Mikhaiel · Aaron Emmanuel · Muhammad Saad Farooq Hamirani · Sheik (Farhaan) Elaheebocus

Course

Robot Motion Planning · Professor Jonathan Kelly