Skip to content

MPC RRT*

Welcome to the documentation for the MPC RRT* Motion Planning Framework. This package provides a modular reference implementation of a deterministic RRT* planner tightly coupled with a bicycle-model MPC controller. The documentation is organised as follows:

  • Architecture – software modules, data flow and configuration boundaries.
  • MPC Theory – optimal control problem formulation and solver tuning strategies.
  • Vehicle Dynamics – single-track model derivation and linearisation steps.
  • RRT* Planning – algorithmic details, heuristics and determinism guarantees.
  • Pipeline Usage – running the end-to-end system from CLI and Python APIs.
  • Pipeline Integration – end-to-end sequencing, data contracts and MPC/RRT* interaction notes.
  • Configuration – YAML schema, defaults and override tips.
  • Developer Guide – contribution workflow and extension points.

Each document is self-contained and includes references to the relevant code modules. All equations are provided using standard Markdown+LaTeX syntax.

Capabilities at a Glance

  • Deterministic planning – reproducible RRT* exploration trees driven by a shared random seed and deterministic collision checking on occupancy grids.
  • Model-predictive control – OSQP-backed MPC with velocity, steering and rate constraints, plus automatic relaxation strategies for infeasible iterations.
  • Visualization toolchain – Matplotlib utilities for RRT* trees, MPC predictions and frame recording that exports to PNG or GIF.
  • Headless automation – CLI entry points and Python APIs that run the full pipeline while persisting artefacts under plots/ for later inspection.

Repository Layout

  • src/ – production code organised by domain (maps, planning, control, pipeline orchestration, visualisation helpers).
  • tests/ – deterministic unit tests covering geometry helpers, planning and MPC building blocks.
  • docs/ – this MkDocs site documenting architecture, algorithms and operator workflows.
  • examples/ – runnable scripts demonstrating pipeline usage and GIF generation.
  • assets/ & plots/ – input maps and generated artefacts respectively.

Execution Modes

  • Command line – python -m src.cli run executes the full stack with live visualisation; additional sub-commands target individual stages.
  • Library – import PipelineOrchestrator or run_pipeline to embed the planner/controller stack inside notebooks, experiments or services.
  • Batch – schedule simulations headlessly by configuring viz.backend: Agg and enabling frame recording for offline review.

Observability

The project configures structured logging via src.logging_setup. Core stages emit INFO-level progress updates (map preparation, planning iteration milestones and MPC roll-out summaries) so terminal output mirrors the high-level timeline of a run. Adjust log levels with logging.basicConfig or the helper configure_logging() to obtain more verbose DEBUG traces during development.