24 March 2024 in Proceedings of the AAAI Conference on Artificial Intelligence
The Spiking Neural Network (SNN), as one of the biologically inspired neural network infrastructures, has drawn increasing attention recently. It adopts binary spike activations to transmit information, thus the multiplications of activations and weights can be substituted by additions, which brings high energy efficiency. However, in the paper, we theoretically and experimentally prove that the binary spike activation map cannot carry enough information, thus causing information loss and resulting in accuracy decreasing. To handle the problem, we propose a ternary spike neuron to transmit information. The ternary spike neuron can also enjoy the event-driven and multiplication-free operation advantages of the binary spike neuron but will boost the information capacity. Furthermore, we also embed a trainable factor in the ternary spike neuron to learn the suitable spike amplitude, thus our SNN will adopt different spike amplitudes along layers, which can better suit the phenomenon that the membrane potential distributions are different along layers. To retain the efficiency of the vanilla ternary spike, the trainable ternary spike SNN will be converted to a standard one again via a re-parameterization technique in the inference. Extensive experiments with several popular network structures over static and dynamic datasets show that the ternary spike can consistently outperform state-of-the-art methods. Our code is open-sourced at https://github.com/yfguo91/Ternary-Spike.
1 December 2018 in Measurement
10 November 2017 in IEEE Sensors Journal
Obtaining the land vehicle’s position, azimuth and attitudes autonomously and accurately is significant for vehicle based weapons’ combat effectiveness. High performance INS can provide high precision attitudes and azimuth reference information, however, it is usually very expensive. In addition, due to gyroscope drifts and accelerometer biases, INS navigation errors would diverge with time. To solve the problem, this paper proposed an integrated positioning and orientation method based on the FOG single-axis rotational INS (FRINS) and the odometer (OD). The proposed method adopted the rotation modulation technique to suppress the inertial sensor errors, particularly z-axis FOG drift, by which the navigation accuracy, especially the azimuth accuracy, can be improved. At the same time, the proposed method integrated the FRINS with odometer to suppress INS errors’ divergence with time. Combining the above rotation modulation technique and the integration of FRINS and OD, the proposed method could improve the integrated system positioning and orientation accuracy as well as reduce the integrated system cost. This paper presented the FRINS/OD integrated navigation system configuration, established the FRINS/OD integrated system errors model, and then verified the performance of proposed method by simulations. The results show that the proposed method could achieve positioning accuracy of 0.03% (of total traveled distance) and orientation of ±20″ respectively using RINS with FOGs of 0.1°/h and accelerometers of 500ug.
10 December 2021 in IEEE Robotics and Automation Letters
Active simultaneous localization and mapping (SLAM) is an important technique for mobile robots to autonomously explore and map an environment. This letter considers the problem of active SLAM with a prior topo-metric graph. Unlike existing works, we consider a more challenging scenario that there exist indistinguishable subgraphs in the prior graph and the robot does not know its initial position. In this scenario, the robot cannot plan a global path for mapping at the very beginning. To address this problem, we propose a novel framework which consists of two stages, active localization and active mapping. Our main contributions include a graph matching approach for localization and an active localization strategy. The graph matching approach is designed to reliably localize the ego position in the prior graph, by which the switch from active localization to active mapping can be determined reliably. The active localization strategy uses only the maximum likelihood state as input to reduce the computational complexity of motion policy generation. Experimental results demonstrate that our active SLAM system is able to reliably identify its ego position in the prior graph and map an environment in the presence of indistinguishable subgraphs.
1 January 2024 in IEEE Signal Processing Letters
Recently, the neuromorphic vision sensor has received more and more interest. However, the neuromorphic data consists of asynchronous event spikes, which makes it difficult to construct a big benchmark to train a power general neural network model, thus limiting the neuromorphic data understanding for “unseen” objects by deep learning. While for the frame image, since the training data can be obtained easily, the zero-shot and few-shot learning for “unseen” task via the large Contrastive Vision-Language Pre-training (CLIP) model, which is pre-trained by large-scale image-text pairs in 2D, have shown inspirational performance. We wonder whether the CLIP could be transferred to neuromorphic data recognition to handle the “unseen” problem. To this end, we materialize this idea with NeuroCLIP in the paper. The NeuroCLIP consists of 2D CLIP and two specially designed modules for neuromorphic data understanding. First, an event-frame module that could convert the event spikes to the sequential frame image with a simple discrimination strategy. Second, an inter-timestep adapter, which is a simple fine-tuned adapter based on a spiking neural network (SNN) for the sequential features coming from the visual encoder of CLIP to improve the few-shot performance. Various experiments on neuromorphic datasets including N-MNIST, CIFAR10-DVS, and ES-ImageNet demonstrate the effectiveness of NeuroCLIP.