24 September 2025 in IEEE Internet of Things Journal
In recent years, artificial intelligence (AI) technologies represented by foundation models have experienced rapid development. Concurrently, on-device AI inference has become the primary approach for intelligent technology applications, offering advantages such as low latency, high security, and personalization. However, due to the limited resources of on-device systems, on-device AI inference faces new challenges, including improving computational efficiency, optimizing task parallelism, and model optimization. This survey addresses these challenges from a software and algorithmic perspective, focusing on three key areas—operator computation: explore methods to accelerate matrix multiplication and convolution, as well as techniques like operator fusion and vectorized computation; task inference: analyze heterogeneous and distributed computing, memory allocation, and energy-efficient tuning to improve the parallel execution and energy efficiency of inference tasks; and AI models: cover model compression, lookup table (LUT) quantization, and model architecture design to reduce computational complexity and storage requirements. By analyzing these areas, the survey aims to improve inference speed, reduce resource dependency, and provide insights into the future trends of on-device AI technology.
30 July 2024 in IEEE Transactions on Automation Science and Engineering
Polar body biopsy has been widely applied in preimplantation genetic diagnosis for assisted reproductive technology. The key step in the polar body biopsy is picking up the polar body from the oocyte/embryo using a micropipette. Unfortunately, the almost transparent appearance of the polar body as well as its dynamic drift when the micropipette approaches it inside the cell makes it a challenging task to pick it up with less cytoplasm loss for the cell. The unnecessary cytoplasm loss in the picking up process of the polar body easily causes damage to the development competence of the cell and may lead to disturbances to the biopsy results of the polar body. This paper proposes a precise robotic picking up method of polar bodies with less cytoplasm loss for biopsy purposes. First, a defocus imaging method is proposed to locate the polar body with an almost transparent appearance. Then, the dynamic drift of the polar body with the micropipette moving inside the cell is modeled online based on force analysis to determine an appropriate trajectory for the micropipette to approach the polar body. Further, an Active Disturbance Rejection Controller (ADRC) is designed to move the micropipette along the desired trajectory to approach the polar body and then aspirate it into the micropipette. The experimental results on porcine oocytes demonstrate that our system is capable of localizing the polar body with a success rate of 95 $\%$ and an average error of 1.12 $\pm$ 0.14 $\mu$ m. Moving along the determined trajectory, the micropipette is capable of approaching the edge of the polar body with an average error of 1.84 $\pm$ 0.31 $\mu$ m (n $=$ 20), which is only 11 $\%$ of the results obtained without dynamic drift estimation of the polar body. With this advantage, our system picks up the polar body with a close 60 $\%$ improvement in success rate (95 $\%$ vs 60 $\%$ ) and only half of the average cytoplasm loss (5 $\%$ vs 10 $\%$ ) in comparison to operation results without dynamic drift estimation. Note to Practitioners —Picking up of polar body from the oocyte/embryo using a micropipette is a vital operation in the polar body biopsy. Precisely picking up the polar body with less cytoplasm is vital to maintaining the developmental competence of the embryo/oocyte and reducing disturbances to biopsy result. This article presented a precise robotic picking up process of polar body. This process introduced defocus imaging method for polar body localization, dynamic drift estimation of polar body, and micropipette trajectory design and motion control by Active Disturbance Rejection Controller (ADRC). Experimental results have demonstrated the efficiency of the proposed robotic picking up process. Application of this process may provide an economical and practical method to carry polar body biopsy for practitioners.
3 October 2024 in IEEE Transactions on Biomedical Engineering
Objective: A robotic fast dual-arm patch clamp system with controllable mechanical stimulation is proposed in this paper for mechanosensitive excitability research of neurons in brain slice. Methods: First, a kinematic model of a dual-arm patch clamp system combined with Monte Carlo method is developed to calculate the workspaces of recording micropipette and stimulation micropipette, and optimize the length of end effector for reducing collision incidences during operation. Then, a quantitative stimulation method to cells using one micropipette is developed based on pressing depth control. Finally, a fast robotic dual-arm patch clamp operation process is proposed based on a three-stage motion control of dual micropipettes to approach target cells and form whole-cell recording with quantitative mechanical stimulation. Results: Experimental results on 50 pyramidal neurons in the primary visual cortex of mouse brain slices demonstrate that this system achieves a threefold throughput with a 37% improvement in the success rate of the contact process and a 42% improvement in the success rate of whole-cell recording in comparison to manual operation. With these advantages, a mechanical stimulation-regulated increase in neuron excitability is observed in primary visual cortex. The experimental results also show that the sodium ion current may be more sensitive to mechanical stimulation than potassium ion current. Conclusion: Our system significantly improves the efficiency of mechanical stimulation induced excitability research of neurons in brain slices. Significance: Our methods have the potential to investigate pathological and pathogenic mechanisms of mechanosensitive ion channel dysfunction-induced diseases in the future.
18 December 2023 in IEEE Transactions on Automation Science and Engineering
Patch clamp technique is a ”gold standard” approach to the study of ion channel biophysics and pharmacology. Due to the complexity of the operation and the heavy reliance on the experience of the experimenter, more and more researchers are focusing on patch clamping automation. In manual patch clamping, the operators patch the neurons in the specific brain region by experience, which lacks quantitative indicators of brain region selection. At present, the existing automated patch clamp focuses more on searching and measuring single cells under a high magnification lens, but ignores the localization of different brain regions under a low magnification lens, affecting the efficiency and application of the automated system. Addressing this, a 2D-3D registration method is proposed to register the experimental microscopic image of brain slices onto the brain atlases. This method is then embedded into the patch clamp system for brain region selection and electrophysiological recordings of neurons in multiple brain regions. It is the first time to achieve automated selection of brain regions in patch clamp experiments. The experimental results indicate that brain slice registration and brain region selection completely meet the requirements of the patch clamp experiment. Compared with the original robotic patch clamp process, the total time for the whole-cell patch clamp initialization with the improved process is reduced by 280 seconds, and the efficiency is increased by 42%. Meanwhile, the time for brain region switching is reduced to $1/6$ of the original. In the future, this method will be applied to automatic image-guided patch clamp systems. Note to Practitioners —Calibrating the depth of the experimental brain slice in the brain and dividing the distribution of each brain region in the brain slice are the most important preparations for electrophysiological experiments. In this paper, we design 2D-3D registration to establish a method for brain region selection and robotic patch clamping of different brain regions in one brain slice. We have successfully detected the electrophysiological signals of neurons in 3 major brain regions, including the visual region, auditory region, and hippocampus. The experimental results show that the method proposed in this paper is suitable for the existing robotic patch clamp system and greatly improves the degree of automation.