Abstract: The developments in data-based, digital artificial intelligence and learning capabilities in computation, algorithms and cognition have tremendously grown in the past decades, while the development of robots’ bodies, morphology and materials has lagged behind. This keynote addresses this challenge and how computational aspects of implementing robotic and computer vision algorithms are currently addressed to deal with multi-sensor data with spatial probabilistic distributions. The current digital tools and simulators are convenient and practical for exploring the quantitative behavior of specific computing neural networks, but their performance is largely dependent of supercomputing capacities. Even the largest supercomputing systems to date are not capable of obtaining real-time performance when running simulations large enough to accommodate multiple areas and layers of computing. Custom digital systems that exploit parallel graphical processing units (GPUs), field programmable gate arrays (FPGAs) or memristors offer good capabilities in computer efficiency, and resilience. In this talk we present the design of a memristor based hybrid-in memory processing architecture of computer vision. The design includes circuitry that controls and enables the in-memory processing of arithmetic operations through memristor cells, sense integrators and other peripheries in order to perform the needed modules for the implementation of the algorithm. In the talk we also address our current attempt to implement efficient “haze removal” CNN to remove the distortion from underwater images. Restoring the underwater images without distortion is important since it increases the quality and performance of the machine learning programs which improves the marine robotics.