IMU prediction validation

[koide3]

Validate if IMU-based state prediction is good or not.
This would be useful to find a mis-configured T_lidar_imu.

[Copilot]

Pull request overview

Adds runtime validation to assess whether IMU-based state prediction is outperforming a simple constant-velocity (no-IMU) baseline, helping diagnose issues such as misconfigured T_lidar_imu or poor bias estimates.

Changes:

• Introduces a new IMUValidation utility that tracks and periodically logs prediction/bias quality statistics.
• Wires IMU validation into the IMU odometry pipeline and exposes a new validate_imu config flag.
• Updates default odometry configs and build system to include the new feature.

Reviewed changes

Copilot reviewed 7 out of 7 changed files in this pull request and generated 3 comments.

Show a summary per file

File	Description
src/glim/odometry/odometry_estimation_imu.cpp	Instantiates and calls the new IMU validation during frame insertion.
src/glim/common/imu_validation.cpp	Implements prediction-vs-baseline and bias validation logic + periodic logging.
include/glim/odometry/odometry_estimation_imu.hpp	Adds validate_imu param and IMUValidation member.
include/glim/common/imu_validation.hpp	Declares the IMUValidation helper class.
config/config_odometry_gpu.json	Documents and enables the validate_imu option.
config/config_odometry_cpu.json	Enables validate_imu option (but with inconsistent section labeling).
CMakeLists.txt	Adds imu_validation.cpp to the glim library build.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[Copilot]

IMU validation is executed unconditionally, but when num_imu_integrated < 2 the code overwrites predicted_T_world_imu/predicted_v_world_imu with a non-IMU-based fallback (or may have an underconstrained IMU factor). That makes the validation results misleading and can trigger incorrect warnings about T_lidar_imu/bias. Consider skipping imu_validation->validate(...) when there isn't sufficient IMU data (e.g., num_imu_integrated < 2), or pass a flag indicating whether the prediction truly came from IMU integration.

Suggested change

	imu_validation->validate(
	Eigen::Isometry3d(last_T_world_imu.matrix()),
	last_v_world_imu,
	Eigen::Isometry3d(predicted_T_world_imu.matrix()),
	predicted_v_world_imu,
	new_frame->T_world_imu,
	new_frame->v_world_imu,
	new_frame->stamp - last_stamp);
	imu_validation->validate(new_frame->imu_bias);
	// Only run IMU validation when the prediction is actually based on IMU integration.
	// For example, when num_imu_integrated < 2, predicted_T_world_imu/predicted_v_world_imu
	// may come from a non-IMU fallback or an underconstrained factor, which would make
	// the validation misleading.
	if (num_imu_integrated >= 2) {
	imu_validation->validate(
	Eigen::Isometry3d(last_T_world_imu.matrix()),
	last_v_world_imu,
	Eigen::Isometry3d(predicted_T_world_imu.matrix()),
	predicted_v_world_imu,
	new_frame->T_world_imu,
	new_frame->v_world_imu,
	new_frame->stamp - last_stamp);
	imu_validation->validate(new_frame->imu_bias);
	}

[Copilot]

config/config_odometry_cpu.json places the new validate_imu setting under a repeated // Optimizer params section, but the parameter is described as a logging/misc flag in the GPU config and in code (// Logging params). This inconsistency makes the config harder to understand/search. Consider moving validate_imu (and save_imu_rate_trajectory / num_threads if intended) under a consistent // MISC/// Logging section and/or updating the header comment block to document it like the GPU config does.