Previous work utilized three-dimensional (3-D) convolutional neural networks (CNNs) tomodel the spatial appearance and temporal evolution concurrently for sign language recognition (SLR) and exhibited impressive performance. However, there are still challenges for 3-D CNN-based methods. First, motion information plays a more significant role than spatial content in sign language. Therefore, it is still questionable whether to treat space and time equally and model them jointly by heavy 3-D convolutions in a unified approach. Second, because of the interference from the highly redundant information in sign videos, it is still nontrivial to effectively extract discriminative spatiotemporal features related to sign language. In this study, deep R(2+1)D was adopted for separate spatial and temporal modeling and demonstrated that decomposing 3-D convolution filters into independent spatial and temporal convolutions facilitates the optimization process in SLR. A lightweight spatial–temporal–channel attention module, including two submodules called channel–temporal attention and spatial–temporal attention, was proposed to make the network concentrate on the significant information along spatial, temporal, and channel dimensions by combining squeeze and excitation attention with self-attention. By embedding this module into R(2+1)D, superior or comparable results to the state-of-the-art methods on the CSL-500, Jester, and EgoGesture datasets were obtained, which demonstrated the effectiveness of the proposed method.

A novel ultra wideband (UWB)‐based scheme is proposed to provide robust and accurate robot localisation in indoor environments. An extended Kalman filter (EKF), which is suboptimal, is combined in the main estimator design with an extended unbiased finite impulse response (EFIR) filter, which has better robustness. In the integrated EKF/EFIR algorithm, the EFIR filter and the EKF operate in parallel and the final estimate is obtained by fusing the outputs of both filters using probabilistic weights. Accordingly, the EKF/EFIR filter output ranges close to the most accurate one of the EKF and EFIR filters. Experimental testing has shown that the EKF/EFIR‐based UWB‐range robot localisation is more robust than the EKF‐ and EFIR‐based ones in uncertain noise environments.

In this article, we presents an effective semantic grounding scheme to long-term mobile robots for dynamic object search in open/dynamic home environments. The challenge of this strategy lies in dealing with situations where the robot suffers the home environment with an unknown layout of the object and the target object is moved without the robot’s knowledge. Therefore, a general probabilistic model and a semantic model are established by investigating typical spatial location relations between object and room type, in order to guide the robot to prioritize the search effort for spaces that are most promising to find the target object. Furthermore, a semantic grounding solution is proposed to address the mutual independence between objects, endowing the robot with inference capability for dynamic object search. Also, our proposal can be shared with other robots that have not explored the environment in order to perform object search tasks. The presented scheme is fully evaluated by extensive experiments in the real-world environment. The results demonstrate the validity of our approach by comparing it with the other three schemes in terms of search time and trajectory length, and show that the proposed method allows the mobile robot to efficiently and robustly find the dynamic object while achieving the human-like performance.