Multi-modal reinforcement learning (RL) has been brought into focus due to its ability to provide complementary information from different sensors, enriching observations of agents. However, the introduction of multi-modal highdimensional observations brings challenges to sample efficiency. There is a lack of research on how to efficiently obtain multi-modal latent states while encouraging them to generate complementary information. To address this, we propose a representation learning method, Multi-modal Joint Predictive Representation (MJPR), which utilizes multi-modal interactive information to predict future latent states. The joint prediction method achieves the representation training for modalities and promotes each modality to generate complementary information related to predictions of each other. In addition, we introduce multi-modal loss balancing to prompt training equilibrium and cross-modal contrastive learning (CMCL) to align the modalities for effective modal interaction. We establish the multi-modal environments in the Deepmind Control suite (DMC) and Webots and compare our method with current RL representation methods. Experimental results show that MJPR outperforms state-of-the-art methods by an average of 12.0% on six subtasks in DMC environments. It outperforms advanced methods by 16.7% and 55.4% in simple tasks and complex tasks of Webots environment, respectively. Moreover, ablation experiments are established in the DMC environment to verify the importance of each module to MJPR.

With the increasing demands for unmanned aerial vehicle (UAV)-based autonomous inspections in the oil and gas industry, one of the challenging issues for 3-D UAV positioning has emerged due to the satellite signal blocking. Considering the existing characteristics of the ultrasonic-based technique, such as the low cost, extremely lightweight, and high positioning accuracy, it can be promising as the potential solution. Nevertheless, the low position update rate and vulnerable positioning performance to the changing environment still limit its applications on UAV. Therefore, in this article, an ultrasonic- and inertial measurement unit (IMU)-based localization algorithm and low-cost UAV autonomous inspection system are presented. With the incorporation of the IMU, the position update rate, accuracy, and stability of the algorithm can all be significantly improved. This is done by the adaptively estimated noise covariance matrices through the proposed adaptive extended Kalman filter algorithm and the added weighting factors. Followed by, an additional virtual observation process is presented to overcome the unavailability of the observation information for further performance improvement. Finally, extensive numerical results and field tests demonstrate that the proposed algorithm and system can achieve the high update rate, reliable, accurate, and precision UAV positioning in oil and gas pressure vessels and are feasible for the UAV autonomous inspection in these environments.