RESUMO
The labor shortage and rising costs in the greenhouse industry have driven the development of automation, with the core of autonomous operations being positioning and navigation technology. However, precise positioning in complex greenhouse environments and narrow aisles poses challenges to localization technologies. This study proposes a multi-sensor fusion positioning and navigation robot based on ultra-wideband (UWB), an inertial measurement unit (IMU), odometry (ODOM), and a laser rangefinder (RF). The system introduces a confidence optimization algorithm based on weakening non-line-of-sight (NLOS) for UWB positioning, obtaining calibrated UWB positioning results, which are then used as a baseline to correct the positioning errors generated by the IMU and ODOM. The extended Kalman filter (EKF) algorithm is employed to fuse multi-sensor data. To validate the feasibility of the system, experiments were conducted in a Chinese solar greenhouse. The results show that the proposed NLOS confidence optimization algorithm significantly improves UWB positioning accuracy by 60.05%. At a speed of 0.1 m/s, the root mean square error (RMSE) for lateral deviation is 0.038 m and for course deviation is 4.030°. This study provides a new approach for greenhouse positioning and navigation technology, achieving precise positioning and navigation in complex commercial greenhouse environments and narrow aisles, thereby laying a foundation for the intelligent development of greenhouses.
RESUMO
Excessive invalid explorations at the beginning of training lead deep reinforcement learning process to fall into the risk of overfitting, further resulting in spurious decisions, which obstruct agents in the following states and explorations. This phenomenon is termed primacy bias in online reinforcement learning. This work systematically investigates the primacy bias in online reinforcement learning, discussing the reason for primacy bias, while the characteristic of primacy bias is also analyzed. Besides, to learn a policy generalized to the following states and explorations, we develop an online reinforcement learning framework, termed self-distillation reinforcement learning (SDRL), based on knowledge distillation, allowing the agent to transfer the learned knowledge into a randomly initialized policy at regular intervals, and the new policy network is used to replace the original one in the following training. The core idea for this work is distilling knowledge from the trained policy to another policy can filter biases out, generating a more generalized policy in the learning process. Moreover, to avoid the overfitting of the new policy due to excessive distillations, we add an additional loss in the knowledge distillation process, using L2 regularization to improve the generalization, and the self-imitation mechanism is introduced to accelerate the learning on the current experiences. The results of several experiments in DMC and Atari 100k suggest the proposal has the ability to eliminate primacy bias for reinforcement learning methods, and the policy after knowledge distillation can urge agents to get higher scores more quickly.