Detection and distance measurement using sensors is not always accurate. Sensor fusion makes up for this shortcoming by reducing inaccuracies. This study, therefore, proposes an extended Kalman filter (EKF) that reflects the distance characteristics of lidar and radar sensors. The sensor characteristics of the lidar and radar over distance were analyzed, and a reliability function was designed to extend the Kalman filter to reflect distance characteristics. The accuracy of position estimation was improved by identifying the sensor errors according to distance. Experiments were conducted using real vehicles, and a comparative experiment was done combining sensor fusion using a fuzzy, adaptive measure noise and Kalman filter. Experimental results showed that the study’s method produced accurate distance estimations.

The problem with lidar placement is finding a lidar position that reduces the dead zone and improves the point cloud resolution. A lidar placement method is thus proposed for the multiple lidar system of autonomous vehicles to resolve this problem. With the introduction of the lidar occupancy grid, the problem is formulated as an optimization problem. A genetic algorithm is subsequently applied to solve the optimization problem. The experimental results using commercial lidars are subsequently presented to show the usefulness of the proposed method.

We propose a new camera-lidar fusion method for road detection where the spherical coordinate transformation is introduced to decrease the gap between the point cloud of 3D lidar data. The camera's color data and the 3D lidar's height data are transformed into the same spherical coordinate, and then input to the convolution neural network for segmentation. Faster segmentation is possible due to the reduced size of input data. To increase the detection accuracy, this modified SegNet expands the receptive field of the network. Using the KITTI dataset, we present the experimental results to show the usefulness of the proposed method.

This paper presents a cooperative object transportation technique using deep reinforcement learning (DRL) based on curricula. Previous studies on object transportation highly depended on complex and intractable controls, such as grasping, pushing, and caging. Recently, DRL-based object transportation techniques have been proposed, which showed improved performance without precise controller design. However, DRL-based techniques not only take a long time to learn their policies but also sometimes fail to learn. It is difficult to learn the policy of DRL by random actions only. Therefore, we propose two curricula for the efficient learning of object transportation: region-growing and single- to multi-robot. During the learning process, the region-growing curriculum gradually extended to a region in which an object was initialized. This step-by-step learning raised the success probability of object transportation by restricting the working area. Multiple robots could easily learn a new policy by exploiting the pre-trained policy of a single robot. This single- to multi-robot curriculum can help robots to learn a transporting method with trial and error. Simulation results are presented to verify the proposed techniques.