Repository navigation
Description
Port gyro odometry and EKF localization to Awkernel and integrate them with the M1 NDT pipeline to evaluate multi-rate state estimation and real-time scheduling under variable computational load.
Scope:
- Gyro odometry using IMU and vehicle velocity inputs
- EKF-based fusion of NDT pose and gyro odometry twist estimates
- Timestamp alignment, coordinate transforms, and covariance handling
- Use of previously available EKF pose estimates or predictions to initialize subsequent NDT scans
- Replay-based evaluation with concurrent NDT, gyro odometry, and EKF processing
- Measurement of localization accuracy, scheduling delay, deadline misses, end-to-end latency, and input data age
Completion criteria:
- NDT, gyro odometry, and EKF operate as an integrated pipeline on Awkernel.
- Output differences from a fixed Autoware reference satisfy predefined tolerances under equivalent inputs and configurations.
- Task periods, deadlines, buffering policies, and temporal dependencies are documented.
- Scheduling interference from NDT is evaluated under controlled workloads.
- Evaluation results and reproduction steps are recorded.
Out of scope:
Motion distortion correction, live vehicle interfaces, automatic pose initialization, and closed-loop vehicle control.
Existing components:
The following supporting components have already been merged into Awkernel and can be reused for M2:
- IMU driver and vehicle velocity converter: feat(app): add autoware module (imu_driver and vehicle_velocity_converter) #681
- IMU corrector: fix(app): add autoware module (imu_corrector) #687
- Core gyro odometry implementation: feat(app): add autoware module (gyro_odometer) #688; remaining work is tracked in Port Gyro Odometry to Awkernel #754.
Their integration and validation against the M2 interface conventions remain part of this milestone. The presence of these components does not by itself establish completion of the integrated localization pipeline.
Metadata
Metadata
Assignees
Labels
No labels
Activity