Skip to content

feat: introduce autonomous 3D navigation and high-fidelity quadrotor flight stack - #312

Open
AryanXCode646 wants to merge 18 commits into
StellarResearch:mainfrom
AryanXCode646:feat/arl-autonomous-navigation
Open

AryanXCode646 wants to merge 18 commits into
StellarResearch:mainfrom
AryanXCode646:feat/arl-autonomous-navigation

Conversation

@AryanXCode646

Copy link
Copy Markdown
Contributor

Summary

This PR introduces a major expansion of AdaptiveRL, extending the existing reinforcement-learning drone simulator into a more comprehensive autonomous 3D navigation and quadrotor flight simulation platform.

It integrates high-fidelity 6-DOF rigid-body dynamics, classical flight controllers, deterministic path planning, dynamic obstacle avoidance, sensor estimation, flight recording, validation scenarios, and enhanced 3D visualization.

The objective is to establish a modular and reproducible foundation for researching autonomous drone navigation and robust flight control.

Key Changes

1. High-Fidelity 6-DOF Quadrotor Dynamics

  • Enhanced rigid-body flight dynamics with quaternion-based attitude representation.
  • RK4 numerical integration and rotational dynamics.
  • Individual motor thrust modelling and actuator response.
  • Aerodynamic effects, wind disturbances, and ground-effect modelling.
  • Improved collision detection to reduce high-speed tunnelling.

2. Autonomous 3D Navigation

  • Introduced deterministic 3D A* path planning.
  • Added obstacle inflation, collision-aware path validation, and line-of-sight path simplification.
  • Implemented dynamic obstacle modelling and trajectory prediction.
  • Added multi-waypoint mission management and reactive replanning.
  • Introduced autonomous Return-to-Home mission behaviour.

3. Flight Control Systems

  • Added cascaded PID-based quadrotor control.
  • Introduced an LQR controller using linearized quadrotor dynamics.
  • Added analytical motor allocation and actuator constraints.
  • Established classical control baselines for comparison with reinforcement-learning policies.

4. Sensor Modelling and State Estimation

  • Added a six-axis IMU model with measurement noise and bias drift.
  • Introduced complementary attitude estimation using SO(3) rotation mathematics.
  • Improved the sensor and state-estimation interfaces.

5. Flight Testing and Benchmarking

  • Added structured flight recording and telemetry infrastructure.
  • Introduced reproducible autonomous flight scenarios.
  • Added flight performance metrics, outcome classification, and validation reporting.
  • Improved benchmark provenance and checkpoint integrity handling.
  • Added domain randomization support for evaluating robustness under varying physical conditions.

6. Visualization and Flight Deck

  • Enhanced the 3D quadrotor representation and flight visualization.
  • Improved trajectory playback and telemetry inspection.
  • Added configurable destination coordinates and pre-flight target validation.
  • Integrated navigation, attitude, velocity, and obstacle information into the visualization workflow.
  • Updated the Streamlit application and project showcase.

7. Documentation and Testing

  • Expanded architecture and flight-dynamics documentation.
  • Added autonomous navigation and flight validation specifications.
  • Added tests covering dynamics, navigation, controllers, sensors, fault injection, recording, benchmarking, and visualization.
  • Updated CLI and configuration integration.

Engineering Scope and Limitations

This project is an educational and research-oriented simulation platform.

The 6-DOF dynamics and control systems are mathematical simulation models. They should not be interpreted as a validated digital twin or flight-qualified autopilot.

No claim of real-world flight readiness, hardware-in-the-loop validation, or physical drone certification is made by this PR.

Validation

The branch includes dedicated automated tests for the newly introduced and modified components.

Please evaluate the pull request using the repository's CI workflows and review the complete changes against the target main branch before merging.

Expected Outcome

This contribution establishes the foundation for a more modular autonomous flight research platform, bringing together reinforcement learning, classical flight control, motion planning, dynamic obstacle avoidance, sensor estimation, and reproducible simulation-based validation.

- Synchronize visualizer dimensions directly with DroneDynamics6DOF physics
- Add volumetric quadrotor airframe with rotor identification and nose indicator
- Support translucent boundary walls and solid 3D arena floor
- Rotate LiDAR rays in visualizer when body_frame_lidar is active
- Differentiate 5 canonical outcomes: SUCCESS, COLLISION, TIMEOUT, MAX_STEPS, UNKNOWN
- Implement SHA-256 run configuration fingerprinting to invalidate stale simulation state
- Standardize versioned checkpoint metadata (schema_version 1.0.0) with conflict detection
- Add full 6-DOF SAC and PPO evaluation in Tab 3 of Flight Deck
…testing, reference PID controller, and reproducible flight recorder
…lity dynamics, IMU estimation, and benchmark integrity
…ic obstacle avoidance, physics integration, and complete provenance
…anning, and provenance

- Eliminate mutable dynamics state aliasing and calculate genuine inertial acceleration in IMU telemetry
- Enforce authoritative mission completion requiring sequential waypoint traversal, position settling, velocity settling, and collision freedom
- Preserve exact finite IEEE 754 precision in canonical JSON dumps and comprehensive scenario provenance hashing
- Implement continuous swept-sphere dynamic obstacle collision detection with adaptive temporal subintervals
- Implement closed-loop state-estimation-in-the-loop control contract and dedicated scenario
- Enable arbitrary non-grid-aligned 3D A* start/goal coordinates with collision-free neighbor connection and validation
- Couple stage-wise continuous first-order actuator lag across RK4 evaluation stages eliminating integration bias
- Add full test suites covering all residual defect requirements with 528 passing tests
… dynamic collision, complete provenance inference, and branch CI
…rovenance, and enforce provenance consistency
…rena bounds, complete recursive provenance, and verify 6-DOF continuous collision
…d static/boundary safety, and recursive provenance
…ng, defensible acceleration, and recursive descriptor provenance
fix(phase3): harden LQR allocation, provably collision-free A*, dynamic obstacle avoidance, physics integration, and complete provenance

This branch has not been deployed

No deployments
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant