Two papers from Intelligent Walker were selected for the top international conference ICRA2023

Two papers published by the VIGGO were accepted by the top international conference ICRA 2023, namely "Faster-LIO: Lightweight Tightly Coupled Lidar-inertial Odometry using Parallel Sparse Incremental Voxels" and Anderson Acceleration for on-Manifold Iterated Error State Kalman Filters." The two papers address the difficult problems in automated driving on the ground and propose an efficient LiDAR inertial odometry based on incremental voxels and an iterated extended Kalman filter based on Anderson acceleration, respectively.

IEEE InternationalConference on Robotics and Automation (IEEE ICRA) is organized by the IEEE International Robotics and Automation Association once a year, and IEEE ICRA is the top international conference in the field of robotics, ranking first in terms of scale (more than 1,000 people) and influence. IEEE ICRA is the top international conference in robotics in terms of size (over 1,000 attendees) and influence. It is the premier international forum for authoritative researchers in robotics to present their findings.

Faster-LIO: Lightweight, Tightly Coupled Lidar-inertial Odometry using Parallel Sparse Incremental Voxels

This paper presents an incremental voxel-based lidar-inertial odometry (LIO) method for fast-tracking rotating and solid-state lidar scanning point clouds. To obtain a high tracking speed, this paper neither uses a complex tree-based structure to partition the spatial point cloud nor a strict k-nearest neighbor (k-NN) query to compute the point matches.

The incremental voxel (iVox) is used as the point cloud spatial data structure, which is modified from the traditional voxel to support incremental insertion and parallel approximate k-NN query, and this paper proposes the linear iVox and the PHC (pseudo-Hilbert curve) iVox as two optional underlying structures in the algorithm.

Experiments show that the iVox achieves speeds of 1000-2000 HZ per scan in solid-state lidar and over 200 Hz in 32-line rotating lidar (using only modern CPUs) while maintaining the same accuracy level!

Anderson Acceleration for on-Manifold Iterated Error State Kalman Filters

In this paper, it is shown that the Iterated Extended Kalman Filter (IEKF) is an estimator that is widely used in real-time localization applications to save computational resources by iterating the observation equations several times to find better linearization points, and each time, it only performs state estimation at the current point in time.

Inspired by the recent iterative closest point algorithm (iterative closest point algorithm), this paper proposes an accelerated iterative error state Kalman filter (IESKFs). The results show that the IESKF can be converted into an iterative immobilization problem, which allows the Anderson acceleration (AA) to be directly applied to the iteration of the IESKF since the error states of the IESKF naturally exist in the tangent space without the need for additional transformations.

However, the currently estimated tangent space may change during the iteration period, so this paper should transform the tangent space to the starting point to perform the Anderson acceleration. The article proposes the AA-IEKF and applies it to a lidar inertial odometry (LIO) system to estimate its proper motion. Experiments show that the Anderson acceleration can effectively reduce the number of iterations of the ESKF and decrease the total computation.