With the development of robot technology and the arrival of industry 4.0 era, society pays more attention to col- laboration and interaction between human and robots. However, safety is still main concern in the development of human-robot collaboration. In this paper, a novel real-time collision avoidance approach for mobile manipulator is proposed by considering the motion status of the human, which includes the relative minimum distance and velocity (both magnitude and direction) between the robot and the human. The distance and velocity of the human hand are first estimated online using a vision sensor, and then defined as danger factors in the potential function of the potential field. The novel potential function proposed in this paper considers not only the safety problem, but also the efficient problem, i.e., the manipulator can make smart control decision to avoid the collision according to the relative velocity in case of the cross over. To overcome the local minimum problem and choose a best motion direction, we propose a motion sampling mechanism for motion planning. For each sample, the robot calculates the potential function to evaluate the safety and efficiency, and chooses a direction which is best for avoidance. We finally demonstrate our idea on a real mobile manipulator platform in a simulated co-worker environment.

In dynamic environment, the suddenly appeared human or other moving obstacles can affect the safety of the bridge crane. For such dangerous situation, the bridge crane must predict potential collisions between the payload and the obstacle, keep safe distance while the swing of the payload must be considered in the mean time. Therefore, the safe distance is not a constant value, which must be adaptive to the relative speed of the bridge crane. However, as far as we know, the mathematical model between the safe distance and the relative speed of the bridge crane has never been fully discussed. In this paper, we propose a safe distance prediction method using model prediction control (MPC), which can make sure that the crane can stop before the obstacle, and avoid possible collisions, while the relative speed and anti-swing are both considered. The experimental results prove the effectiveness of our idea.

In this paper, we propose a distributed solution for3D human pose estimation using a RGBD camera network. Thekey feature of our method is a dynamic hybrid consensus filter(DHCF) is introduced to fuse the multiple view informationof cameras. In contrast to the centralized fusion solution,the DHCF algorithm can be used in a distributed network,which requires no central information fusion center. Therefore,the DHCF based fusion algorithm can benefit from manyadvantages of distributed network. We also show that theproposed fusion algorithm can handle the occlusion problemseffectively, and achieve higher action recognition rate comparedto the ones using only single view information.

Full resolution depth is required in many realworld engineering applications. However, exist depth sensorsonly offer sparse depth sample points with limited resolutionand noise, e.g., LiDARs. We here propose a deep learningbased full resolution depth recovery method from monocularimages and corresponding sparse depth measurements of targetenvironment. The novelty of our idea is that the structure similarinformation between the RGB image and depth image is used torefine the dense depth estimation result. This important similarstructure information can be found using a correlation layerin the regression neural network. We show that the proposedmethod can achieve higher estimation accuracy compared tothe state of the art methods. The experiments conducted on theNYU Depth V2 prove the novelty of our idea.

Visual simultaneous localization and mapping (SLAM) is the core of intelligent robot navigation system. Many traditional SLAM algorithms assume that the scene is static. When a dynamic object appears in the environment, the accuracy of visual SLAM can degrade due to the interference of dynamic features of moving objects. This strong hypothesis limits the SLAM applications for service robot or driverless car intherealdynamicenvironment.Inthispaper,adynamicobject removal algorithm that combines object recognition and optical flow techniques is proposed in the visual SLAM framework for dynamic scenes. The experimental results show that our new method can detect moving object effectively and improve the SLAM performance compared to the state of the art methods.

In this paper, we propose a deep neural networkthat can estimate camera poses and reconstruct thefull resolution depths of the environment simultaneously usingonly monocular consecutive images. In contrast to traditionalmonocular visual odometry methods, which cannot estimatescaled depths, we here demonstrate the recovery of the scaleinformation using a sparse depth image as a supervision signalin the training step. In addition, based on the scaled depth,the relative poses between consecutive images can be estimatedusing the proposed deep neural network. Another novelty liesin the deployment of view synthesis, which can synthesize anew image of the scene from a different view (camera pose)given an input image. The view synthesis is the core techniqueused for constructing a loss function for the proposed neuralnetwork, which requires the knowledge of the predicted depthsand relative poses, such that the proposed method couples thevisual odometry and depth prediction together. In this way,both the estimated poses and the predicted depths from theneural network are scaled using the sparse depth image as thesupervision signal during training. The experimental results onthe KITTI dataset show competitive performance of our methodto handle challenging environments.