Detection and classification of road-objects like cars, pedestrians, and cyclists is the first step in autonomous driving. In particular, point-wise object segmentation for 3D point clouds is essential to estimate the precise appearances of the road-objects. In this paper, we propose SPriorSeg, a fast and accurate point-level object segmentation for point clouds by integrating the strengths of deep convolutional auto-encoder and region growing algorithm. Semantic segmentation using the light-weighted convolutional auto-encoder generates semantic prior by labeling a spherical projection image of point clouds pixel-by-pixel with classes of road-objects. The region growing algorithm achieves pixel-wise instance segmentation by taking into account semantic prior and geometric features between neighboring pixels. We build a well-balanced, pixel-level labeled dataset for all classes using 3D bounding boxes and point clouds from the KITTI object dataset. The dataset is employed to train our light-weighted neural network for semantic segmentation and demonstrate the performance of both semantic and instance segmentation of SPriorSeg.

For driverless driving cars, it is essential to detect drivable space. It can directly apply to plan driving paths by acquiring the occupancy grid map. In addition, it can enhance object clustering by removing the ground in advance. However, in urban, not only a large number of vehicles are driving at the same time, but also roads with diverse inclinations are complicatedly connected with each other. Thus, it is challenging to extract traversable space properly from complex structured environment. For this reason, this paper proposes the real-time drivable space detection for complex urban environment by integrating the model-based segmentation and the region-based segmentation. Moreover, the proposed method utilizes point cloud from 3D LiDAR because it is effective to understand surrounding topography. It is demonstrated using hand-labeled point cloud dataset collected in various types of urban roads by estimating numerical performances and by visualizing results.

This paper proposes a motion planning method for an autonomous vehicle in an urban environment. The proposed method generates trajectories using the closed-loop forward simulation (CLFS), which can consider various properties of an autonomous vehicle. The safeties of the generated trajectories are evaluated using an occupancy grid map. Using the proposed method, smooth and safe trajectories can be generated. The experiment in a real urban environment was conducted to verify the performance of the proposed method.

This paper describes a practical approach to solve a path planning problem in a home environment. The proposed approach incrementally constructs the hierarchical roadmap which has a multi-layered structure using a sonar grid map when a mobile robot navigates in unexplored area. The hierarchical roadmap can almost completely cover the traversable areas in the environment. The mobile robot path planner using the hierarchical roadmap can efficiently search for appropriate paths under the limited computing power and time by reducing the search space size. The benefits of the hierarchical roadmap representation were verified by experiments in a home environment.

This paper proposes an adaptive node sampling method for the probabilistic roadmap (PRM) planner. The proposed method substitutes the random sampling in the learning phase of the PRM planner and improves the configuration of the roadmap. This method uses two phase to determine nodes in order to construct the roadmap. First, the proposed method extracts initial nodes using the approximated cell decomposition and the Harris corner detector. Second, the positions of these nodes are optimized using a construction process of the centroidal voronoi tessellation (CVT). The proposed method determines the adequate number and positions of the nodes to represent the entire free space, and the PRM planner based on the proposed method finds out efficient paths even in narrow passages. These properties have been verified though experiments.

Odometry calibration is a first and essential step to do for a successful navigation because most of control algorithms are based on odomety information. Odometry error can be categorized as systematic and non-systematic error. In this paper, we suggest a novel method to calibrate systematic error using inherent home positioning capability of home cleaning robot. The method is designed for a differential drive type and take advantage of Augmented extended Kalman Fil- ter(AKF) Algorithm to estimates systematic error parameters. Our approach has both characteristics of on-line and off-line. By simulation and experiment, we evaluate the method and the result shows that the proposed method gives odometry error reduction by several times.