XOTAR’s Research and Development focus is highly interdisciplinary, ranging from advanced computer science investigations into massively parallel algorithms and computer architecture for Artificial Intelligence and Cognitive Systems to Distributed Operating Systems, Real-Time Control Systems, Human-Robot Interaction, Robot Perception and Control, and the Psychophysics of Human Signal Detection. The following is a general summary of the areas XOTAR performs R&D:
Human Perceptual Psychophysics – Much of the current research in perceptual psychophysics was done for consumer electronics products such as high fidelity digital audio compression (mp3) for portable music players and high definition television (H.264). Medical devices such as hearing aids are another product that makes use of perceptual psychophysics. XOTAR does research in perceptual psychophysics to engineer Cognitive Systems. Our objective is to understand the structural and functional organization of cortical processes underlying perception to define algorithms, architecture and communication methods used by the human brain to the best extent possible. This R&D requires extending current hypothesizes, and designing experiments with humans including fMRI studies (functional magnetic resonance imaging) and other psychophysical tests that provide further insight into human perception and reasoning for the purposes of modeling and designing intelligent systems with similar capabilities. As XOTAR evolves, the company will reach out to various universities for collaborative investigations into these basic psychophysical phenomena.
Visual System Modeling and Processing - Computational, Psychophysical and Cognitive Research into the active processes involved in acquiring a four dimensional model of and environment for the purpose of guidance, effector manipulation, pursuit/evasion and locomotion is a key investigative area of XOTAR. Our initial focus is the dynamic motor control of eye, head and neck movements used in active perception from a visuomotor and oculomotor standpoint. Early Vision research involves investigations revealing the parallel process used in low level feature extraction, such as texture identification and object segmentation. This research involves discovering the low level operators and adaptive processes used by the human visual system to observe symmetries and invariants in the geometric shape or form of objects that in turn, via learning processes, yield groups that form the basis of object classification, categorization and pattern recognition.
Pattern Recognition – XOTAR’s Active vision System has both passive and active recognition algorithms. Passive algorithms process texture regardless of context or goals, while the active algorithms are contextually and dynamically invoked by high level vision processes to extract specific, targeted information from the scene. XOTAR is doing specific research on passive pattern recognition processes as well as computer software and hardware architecture supporting active or goal driven pattern recognition. Additionally, XOTAR is researching invariant pattern recognition algorithms; which perform independently of occlusions, shadows, luminosity variations or even color and surface texture changes, enabling a higher and more flexible recognition system.
3D Robot Vision – Computer science research into the means of encoding the three dimensional information presented to image sensors and internalizing it into a usable model of information sufficient to support actions such as locomotion and object manipulation is the most fundamental focus of research activity of XOTAR. XOTAR’s 3D Robot vision research investigates both monocular and stereo depth perception.
Geometric Reasoning – Humans are known to use a form of Mental Imagery in both the perceptual and motor planning process of perception-action planning and execution. The robot equivalent is to form a three dimensional geometric model of the environment and localize the observer with respect to it. This model, in addition to a model of the robot itself, is used to project immediate futures and plan movements via geometric transformations that drive motor outputs. XOTAR does extensive research into psychological studies of geometric reasoning in support of the architecture it develops as a core reasoning foundation.
Computational Neurobiology of Human Reaching, Grasping and Pointing – Reaching, Grasping and Pointing are key areas of motor behavior that allow humans to manipulate our environment. Models of how these motor behaviors arise, and the cognitive processes such as motor imagery used to plan movements are an investigative focus of XOTAR.
Wayfinding and Navigation – Many current robots use a form of Simultaneous Localization and Mapping (SLAM) which are algorithms to spatially orient a robot with respect to its environment while simultaneously acquiring a geometric model of the environment. The current techniques developed are largely 2D in nature and support sensing systems that integrate range finding and GPS (Global Positioning System) type input. XOTAR’s systems will perform SLAM as well, with a design that uses active perception to spatially orient and match the robot to an environment, leveraging previous history or referenced data (landmark data acquired from other robots). Additionally, XOTAR’s system will work purely from three dimensional visual processing, which is required for the more complex environments such as interior spaces in homes, offices, factories and other buildings.
Applied Computational Mathematics - XOTAR accesses the full gamut of analytical and computational mathematics used in Advanced Robotics, including emerging applied fields like Computational Topology and Group Theory and modern applied forms of Discrete and Computational Geometry, Manifolds, Differential Geometry, Tensors, and Algebraic Geometry.
