Kan Jiangming, PhD

Forest roads are short of structured terrain. Individual wheels often cannot contact the ground when conventional chassis is driving, and the mobility is weak. In addition, the lateral rollover usually occurs. In this article, a forestry chassis with a novel articulated structure with three degrees of freedom (FC-3DOF(II)) is proposed. Compared with conventional chassis, the novel articulated structure is designed, which contributes to achieving full-time contact between wheels and ground. The mobility is improved. For the lateral stability, the previous lateral rollover model of chassis is often established by the geometrical position of COG (center of gravity) of the frame. This method is applied with limitations, which is not universal. Therefore, a new accurate lateral rollover model for FC-3DOF(II) is derived, which predicts the lateral stability by analyzing tire contact forces. The new lateral rollover model is more general and recovers the previous model. To verify the theoretical analysis exactly, the virtual prototype of FC-3DOF(II) is established in SolidWorks, and simulations of lateral rollover are carried out in ADAMS. In simulation experiments, the lateral stability is predicted by analyzing tire contact forces when the inclination of terrain is increasing. Two conditions are considered in simulations. The lateral stability of FC-3DOF(II) and FC-3DOF(II) installed rectangular objects. Compared to the simulation and theoretical results, for FC-3DOF(II), the maximum absolute percent difference of the contact force with the theoretical analysis relative to the simulation is only 1.83%. For FC-3DOF(II) installed rectangular objects, the simulation results show that the lateral rollover is caused by the rear up-slope wheel when the inclination of terrain reaches 34°. The theoretical result relative to the simulation is only 2.90%. The maximum absolute percent difference of the contact force with the theoretical analysis relative to the simulation is only 2.50%. Simulation results validate the effectiveness of the proposed lateral rollover model in two conditions.

Obstacle-crossing performance is an important criterion for evaluating the power chassis of forestry machinery. In this paper, a new six-swing-arm wheel-legged chassis (SWC&F) is designed according to the characteristics of forest terrain, using herringbone legs to control the ride comfort and stability of the chassis in the process of crossing obstacles. First, the kinematic model of the SWC&F is established, the coordinate analytical expression of each wheel centre position is derived, and the swing angle range of each wheel leg of the chassis is calculated according to the installation position of the hydraulic cylinder. Next, the control model of the system is constructed, and the obstacle-crossing performance of the SWC&F is analyzed by ADAMS/Simulink co-simulation using the PID control method and conventional control method, respectively. The results show that the maximum obstacle crossing height of the SWC&F can reach 411.1 mm, and the chassis with PID control system has good dynamic response characteristics and smooth motion, which meets the requirements of forest chassis obstacle crossing design. The study can provide the foundation for the practical laws of the physical prototype of the forest vehicle chassis.

Visual Place Recognition (VPR) enables robots to determine current location by comparing input image against previously stored reference images. It is essential in autonomous location and simultaneous localization and mapping (SLAM). A key task of VPR is evaluating similarity between images, as state-of-the-art deep learning-based approaches have achieved outstanding performance in standard indoor/outdoor scenes. However, the SOTA deep learning-based methods underperform in forestry robotic owing to two challenges, constrained computational capabilities and appearance variation due to seasonal shifts, weather/light/viewpoint variations, which substantially impair visual similarity computation. Consequently, this work proposes ForestsNet, a novel lightweight VPR network, to resolve this issue. First, a Binary Neural Network (BNN) was constructed to achieve considerable memory reduction. A novel binarization function, Leaky Sign, is proposed; it adaptively applies quantization factors to input activations, it retains richer feature information during binarization while significantly reducing accuracy degradation of place recognition. Second, Mixer Forests, a novel multi-layer perceptron-based aggregation method is introduced to integrate global context into feature maps, substantially enhancing the robustness against appearance variation. In addition, two novel evaluation metrics, Memory Allocation Efficiency and Balance Compression Recall, are designed to quantify the trade-off between memory efficiency and place recognition accuracy. Experimental results demonstrate that ForestsNet achieves substantially higher memory usage efficiency than full-precision networks. Compared to state-of-the-art BNNs, it presents superior performance in both memory efficiency and place recognition accuracy, establishing itself as a robust VPR solution for resource-constrained forestry robots.

Visual Place Recognition (VPR) enables robots to determine current location by comparing input image against previously stored reference images. It is essential in autonomous location and simultaneous localization and mapping (SLAM). A key task of VPR is evaluating similarity between images, as state-of-the-art deep learning-based approaches have achieved outstanding performance in standard indoor/outdoor scenes. However, the SOTA deep learning-based methods underperform in forestry robotic owing to two challenges, constrained computational capabilities and appearance variation due to seasonal shifts, weather/light/viewpoint variations, which substantially impair visual similarity computation. Consequently, this work proposes ForestsNet, a novel lightweight VPR network, to resolve this issue. First, a Binary Neural Network (BNN) was constructed to achieve considerable memory reduction. A novel binarization function, Leaky Sign, is proposed; it adaptively applies quantization factors to input activations, it retains richer feature information during binarization while significantly reducing accuracy degradation of place recognition. Second, Mixer Forests, a novel multi-layer perceptron-based aggregation method is introduced to integrate global context into feature maps, substantially enhancing the robustness against appearance variation. In addition, two novel evaluation metrics, Memory Allocation Efficiency and Balance Compression Recall, are designed to quantify the trade-off between memory efficiency and place recognition accuracy. Experimental results demonstrate that ForestsNet achieves substantially higher memory usage efficiency than full-precision networks. Compared to state-of-the-art BNNs, it presents superior performance in both memory efficiency and place recognition accuracy, establishing itself as a robust VPR solution for resource-constrained forestry robots.