The pavement inspection task, which mainly includes crack and garbage detection, is essential and carried out frequently. The human-based or dedicated system approach for inspection can be easily carried out by integrating with the pavement sweeping machines. This work proposes a deep learning-based pavement inspection framework for self-reconfigurable robot named Panthera. Semantic segmentation framework SegNet was adopted to segment the pavement region from other objects. Deep Convolutional Neural Network (DCNN) based object detection is used to detect and localize pavement defects and garbage. Furthermore, Mobile Mapping System (MMS) was adopted for the geotagging of the defects. The proposed system was implemented and tested with the Panthera robot having NVIDIA GPU cards. The experimental results showed that the proposed technique identifies the pavement defects and litters or garbage detection with high accuracy. The experimental results on the crack and garbage detection are presented. It is found that the proposed technique is suitable for deployment in real-time for garbage detection and, eventually, sweeping or cleaning tasks.

Routine cleaning the pavement is an essential requirement to maintain a sustainable environment for social life. The different width and type of pavements raise the challenges for autonomous vehicles with fixed shape to operate effectively. In this paper, we introduce the vision based reconfiguration of self-reconfigurable pavement sweeping robot called Panthera, which can adjust its frame width to ease the cleaning tasks to become friendly with different pavement geometry. The expansion and compression operations of the Panthera width are implemented by rotating one high torque motor connecting with the lead screw rod to change the opening angle of linkage hinges. The Panthera cleaning and locomotion operations are synchronized with changing the robot width according to the output of detected pavement width. To this end, the segmented pavement leveraged on the masked based deep convolutional neural network (DCNN) is used as input for the proposed closed-loop feedback control method, enabling the robot to adjust the requirement of changing the width during locomotion accurately. The proposed PID scheme takes into account the robot kinematic design with the flexibility of width changing modes. The experiments were carried out in real environments demonstrated the autonomous reconfiguration robot width with various locomotion scenarios on pavements of varying width.

The periodic cleaning pavement is a requirement for a clean environment in urban areas. The robot named Panthera introduced in this paper has the ability to change the width of the frame to help the cleaning tasks become suitable for different pavement types and friendly with the activities of the pedestrian. The Panthera cleaning operations which allow pedestrian walks freely on the pavement is modeled as a pedestrian robot cohabitant framework. To this end, the mask based deep convolutional neural network (DCNN) is used to archive segmented label maps of pedestrians in color image, then the distance from detected objects to the robot is estimated and tracked by averaging filtered depth values in the corresponding region in the refined depth image. The width and the distance from the robot to the approaching pedestrians are used to adjust the robot width. The enlarging and squeezing operations of the Panthera width are conducted by rotating one motor to change the length of the lead screw rod and the angle linkage hinges from the information of encoders sensors. The experiments carried out on real environments demonstrated the autonomous avoiding pedestrians ability by the kinematic model of Panthera.

This paper firstly presents the design and modeling of a quadruped wheeled robot named Tarantula. It has four legs each having four degrees of freedom with a proximal end attached to the trunk and the wheels for locomotion connected at the distal end. The two legs in the front and two at the back are actuated using two motors which are placed inside the trunk for simultaneous abduction or adduction. It is designed to manually reconfigure its topology as per the cross-sections of the drainage system. The bi-directional suspension system is designed using a single damper to prevent the trunk and inside components from shock. Formulation for kinematics of the wheels that is coupled with the kinematics of each leg is presented. We proposed the cost-effective method which is also an on-site approach to estimate the kinematic parameters and the effective trunk dimension after assembly of the quadruped robot using the monocular camera and ArUco markers instead of high-end devices like a laser tracker or coordinate measurement machine. The measurement technique is evaluated experimentally and the same set up was used for trajectory tracking of the Tarantula. The experimental method for the kinematic identification presented here can be easily extended to the other mobile robots with serial architecture designed legs.