Computational and Hardware Models of Active Sensing Behaviors

老鼠有大约60根巨大的面部胡须，它们是非常敏感的触觉传感器。触须的触觉信息被传送到一个叫做三叉神经节的大脑结构，然后再传送到一个叫做三叉神经核的结构。这条通道大致类似于将信息从人体皮肤传递到大脑的通道。哈特曼教授、梅米克教授和奥格伦奇-梅米克教授正在用老鼠的须系统作为模型来研究动物获取和提炼传入的感官信息以构建周围世界表征的过程。该研究在三叉神经核的神经生理学记录和观察到的神经反应的计算模型之间进行迭代。由于神经系统是大规模并行的，这些模型很快就超过了传统计算机所施加的限制。因此，这些模型是用现场可编程门阵列（fpga）实现的，它可以有效地复制脑干电路的巨大并行性。值得注意的是，这些模型是用来处理来自真实和机器人胡须的数据的，并且可以用来预测真实动物的生理反应。这项工作有可能对工业界和学术界产生影响。从工业角度来看，这可能会激发使用非光学传感器的无监督3D物体识别的研究。这种感知能力在各种特殊环境中可能是有用的。例如，视力有限或没有视力的搜索和救援机器人可能会使用类似于这里开发的硬件和算法。在神经生理学中，这项工作揭示了跨模式（视觉、听觉、体感和本体感觉）的处理算法。这项工作还有助于加强博士后、研究生和本科生的跨学科培训，以及少数民族学生的招募。这项工作还补充了PI和co -PI开发的研究生课程。

Rats have approximately sixty large facial whiskers that serve as exquisitely sensitive tactile sensors. Tactile information from the whiskers is carried to a brain structure called the trigeminal ganglion, and then to structures called the trigeminal nuclei. This pathway is roughly analogous to the pathway that carries information from the human skin to the brain. Professors Hartmann, Memik, and Ogrenci -Memik are using the rat whisker system as a model to study the process by which animals acquire and refine incoming sensory information to construct representations of the world around them. The research iterates between neurophysiological recordings in the trigeminal nuclei and computational modeling of the observed neural responses. Because the nervous system is massively parallel, the models quickly exceed the limits imposed by conventional computers. The models are therefore being implemented with Field Programmable Gate Arrays (FPGAs), which can effectively replicate the immense parallelism of brainstem circuitry. Notably, the models are built to process data from both real and robotic whiskers, and can be used to generate predictions about physiological responses in the real animal. This work has the potential to have an impact in industry as well as academia. From an industrial perspective, it is likely to inspire studies on unsupervised 3D object recognition using non-optical sensors. Such sensing ability may be useful in a variety of specialized environments. For example, search and rescue robots operating with limited or no vision could potentially use hardware and algorithms similar to those developed here. In neurophysiology, the work is shedding light on processing algorithms common across modalities (vision, audition, somatosensation and proprioception). The work also contributes to strongly-interdisciplinary training at the post-doc, graduate and undergraduate levels, as well as to minority student recruiting. The work also complements graduate-level courses developed by the PI and Co-PIs.