codetiger
91d669b82f
- Remove bag file recording/playback infrastructure (no longer needed) - Add recently-visited cooldown to prevent frontier re-selection - Add trapped frontier detection to skip unreachable enclosed areas - Increase A* safety margin from 5cm to 10cm for better wall clearance - Update default config: wheel odometry, hybrid_p2l matching, higher velocities - Add navigation config section with robot radius and safety margins - Add sangam-io lib.rs for library usage
SLAM Round-Trip Test Framework
Integration testing for DhruvaSLAM mapping accuracy against mock simulation ground truth.
Quick Start
cd slam-test
# First time: compile protobuf
make proto
# Run with defaults (2 min duration)
make test
# Or run directly with custom options
./test_slam.sh --output ./results/mytest --duration 60
Requirements
- Rust toolchain (for building sangam-io and dhruva-slam)
- Python 3.8+ with packages:
pip install protobuf numpy pillow matplotlib
- protoc (protocol buffer compiler):
# macOS
brew install protobuf
# Linux
apt install protobuf-compiler
Available Commands
make proto # Compile protobuf files (required first time)
make test # Run test (2 min duration)
make test-quick # Run quick test (30 sec)
make test-long # Run long test (5 min)
make clean # Clean results directory
Manual Test Options
./test_slam.sh [OPTIONS]
Options:
--output DIR Output directory (default: results/TIMESTAMP)
--duration SEC Test duration in seconds (default: 120)
--config FILE DhruvaSLAM config file (default: dhruva-test.toml)
--help Show help message
Structure
slam-test/
├── README.md # This file
├── Makefile # Build and test commands
├── test_slam.sh # Main test orchestrator
├── test_config.toml # Test parameters and thresholds
├── dhruva-test.toml # SLAM config (resolution=0.05m)
├── round-trip-test-proposal.md # Detailed design doc
├── scripts/
│ ├── send_command.py # Send commands to DhruvaSLAM
│ ├── analyze_map.py # Compare maps, generate reports
│ └── proto/
│ └── dhruva_pb2.py # Generated protobuf (run make proto)
└── results/ # Test outputs (gitignored)
Test Flow
- Setup: Start SangamIO (mock) and DhruvaSLAM
- Mapping: Send StartMapping, robot explores autonomously
- Capture: Stop mapping, save output map
- Analysis: Compare against ground truth, generate reports
Outputs
Each test run creates a timestamped directory with:
| File | Description |
|---|---|
sangamio.log |
SangamIO process logs |
dhruva.log |
DhruvaSLAM process logs |
ground_truth.pgm |
Input map copy |
slam_output.pgm |
SLAM generated map |
comparison.png |
Visual diff overlay |
analysis.json |
Numeric metrics |
summary.txt |
Human-readable results |
Visual Diff Colors
- White: Free space (both agree)
- Green: Occupied (both agree - correct detection)
- Red: False positive (SLAM occupied, ground truth free)
- Blue: False negative (SLAM free, ground truth occupied)
- Gray: Unknown in SLAM output
Success Criteria
| Metric | Pass | Warning |
|---|---|---|
| Occupied overlap | > 70% | > 60% |
| Free space overlap | > 85% | > 75% |
| Coverage ratio | > 75% | > 60% |
Configuration
Test Config (test_config.toml)
Test parameters and pass/fail thresholds.
SLAM Config (dhruva-test.toml)
DhruvaSLAM configuration with resolution matching ground truth (0.05m = 5cm).
Key settings that must match sangam-io/mock.toml:
[map] resolution = 0.05[lidar]mounting offsets[odometry]encoder parameters
Troubleshooting
"Proto files not compiled"
make proto
"Could not connect to port 5557"
Check if DhruvaSLAM started:
tail -f results/*/dhruva.log
"SangamIO failed to start"
Check if port 5555 is in use:
lsof -i :5555