Robots in the work place can perform hazardous or even 'impossible' tasks; e.g., toxic waste clean-up, desert and space exploration, and more. AI researchers are also interested in the intelligent processing involved in moving about and manipulating objects in the real world.
AI is better at recognizing objects than the average human -- but only under super-specific circumstances. Even a slightly unusual scene can cause it to fail. Why it matters: Image recognition is at the heart of frontier AI products like autonomous cars, delivery drones and facial recognition. But these systems are held back by serious problems interpreting the messy real world.
"People think footballers are all like robots. We can control everything on the pitch. But your heart is beating 200 times a minute. "Jaw-dropping" is a word I deplore. Yet, I came close to muffing a mandible when I encountered stories about a bunch of electronics engineers -- clearly with too much time on their hands -- working feverishly on "the development of robotic soccer players which can beat a human World Cup champion team." According to an explanation of this eccentric dream, "the idea of robots playing soccer was first mentioned by Professor Alan Mackworth" of the University of British Columbia in the early 1990s. Accessing the World RoboCup website, I discovered that a fair amount of money went into studying the "financial feasibility" and "social impact" of lightning-fast, cat-quick robo-soccer teams humiliating the Germans, Brazilians and British hooligans at, say, the 2026 World Cup. The "researchers concluded that the project was feasible and desirable."
A team of researchers at MIT have created a new AI algorithm that can help cameras "see" off-camera things using only moving shadows. In a paper titled "Computational Mirrors: Blind Inverse Light Transport by Deep Matrix Factorization," the scientists at MIT's CSAIL share how they pointed a camera at a pile of objects and then filmed the shadows created on those objects by a person moving around off-camera. The AI analyzed the shadows and was able to reconstruct a blurry but strikingly accurate video of what the person was doing with their hands. While many of the current results may look like a blurry mess of pixels, scientists are working to refine the technology to one day allow cameras to see around corners and other obstructions, something that could be useful in a wide range of applications, from search and rescue to self-driving cars.
"It's fun," says research scientist Janelle Shane of her perpetual learning curve at Boulder Nonlinear Systems, a custom light-control manufacturing company. "This was my first job after my PhD. I knew I wanted to go into industry, and this merges post-doc-style research with business." With her colleagues, Shane works on projects that encompass a multitude of optics-related technologies, from nonmechanical beamsteering for planetary landers and self-driving cars, to ultrafast microscopy and spatial light modulators for neuroscientists. "We're driven by cutting-edge science and pushed to build something new," she says.
Most robots lack the ability to learn new objects from past experiences. To migrate a robot to a new environment one must often completely re-generate the knowledge- base that it is running with. Since in open-ended domains the set of categories to be learned is not predefined, it is not feasible to assume that one can pre-program all object categories required by robots. Therefore, autonomous robots must have the ability to continuously execute learning and recognition in a concurrent and interleaved fashion. This paper proposes an open-ended 3D object recognition system which concurrently learns both the object categories and the statistical features for encoding objects.
Imitation learning has traditionally been applied to learn a single task from demonstrations thereof. The requirement of structured and isolated demonstrations limits the scalability of imitation learning approaches as they are difficult to apply to real-world scenarios, where robots have to be able to execute a multitude of tasks. In this paper, we propose a multi-modal imitation learning framework that is able to segment and imitate skills from unlabelled and unstructured demonstrations by learning skill segmentation and imitation learning jointly. The extensive simulation results indicate that our method can efficiently separate the demonstrations into individual skills and learn to imitate them using a single multi-modal policy. Papers published at the Neural Information Processing Systems Conference.
Fox News Flash top headlines for Dec. 7 are here. Check out what's clicking on Foxnews.com A Tesla on autopilot rear-ended a Connecticut trooper's vehicle early Saturday as the driver was checking on his dog in the back seat, state police said. Police said they had responded to a disabled vehicle that was stopped in the middle of Interstate 95. While waiting for a tow, the self-driving Tesla came down the road.