And how? According to the scientists, that the robots can feel will be achieved thanks to the vision adapted to the context and the tactile sensing in combination with the neuromorphic processing of Intel. ” The fact of bringing the sense of touch to robotics “, they say in the North American company, “can significantly improve capabilities and functionality, compared to current vision-only systems, and how neuromorphic processors can overcome traditional architectures in the processing of these sensory data ”.
This NUS research offers a compelling vision for the future of robotics, in which information is perceived and processed in an event-driven way, combining multiple modalities. The work adds to a growing set of results showing that neuromorphic computing can provide significant improvements in latency and power consumption once the entire system is redesigned in a context-adapted paradigm that encompasses sensors, data formats, algorithms and hardware architecture.
Why is this research important?
The sense of touch is sensitive enough to appreciate the difference between surfaces that differ by a single layer of molecules. However, most of today’s robots operate solely on visual processing. The NUS researchers hope to change this by using the artificial skin they have recently developed. According to their research, it can detect touch more than 1,000 times faster than the human sensory nervous system and identify the shape, texture and hardness of objects 10 times faster than the blink of an eye.
In this sense, allowing a human-like sense of touch in robotics could significantly improve current functionality and even lead to new use cases. For example, robotic arms fitted with artificial skin could easily adapt to changes in factory-made products, using touch sensing to identify and grasp unknown objects with the right amount of pressure to prevent slippage. Likewise, the ability to better feel and perceive the environment could also allow for closer and safer human-robotic interaction, or bring us closer to automating surgical tasks by giving surgical robots the sense of touch that they lack today.
The need for the neuromorphic chip
While creating the artificial skin is a step in bringing this vision to life, it also requires a chip that can draw accurate conclusions based on sensory data from the skin in real time, while operating at a power level efficient enough to to be deployed directly inside the robot. “Creating an ultra-fast artificial skin sensor solves about half the puzzle of making robots smarter,” says Benjamin Tee, associate professor in the Department of Materials Science and Engineering at NUS and the Institute of Innovation and Technology at NUS. the Health of the NUS. “ They also need an artificial brain that can finally achieve perception and learning as another critical piece of the puzzle. Our demonstration of an artificial intelligence skin system with neuromorphic chips like the Intel Loihi is a major step towards energy efficiency and scalability . “
This novel robotic system developed by researchers at the National University of Singapore comprises an artificial brain system that mimics biological neural networks, which can be powered by a low-power neuromorphic processor such as Intel’s Loihi chip, and which is integrated with artificial sensors. skin and vision.
About the investigation
To break new ground in robotic perception, the NUS team began exploring the potential of neuromorphic technology to process sensory data from artificial skin using Intel’s Loihi neuromorphic research chip. In their initial experiment, the researchers used a robotic hand fitted with the artificial skin to read Braille, passing the touch data to Loihi, via the cloud, to convert the micro bumps the hand felt into semantic meaning. In doing so, Loihi achieved more than 92% accuracy in sorting Braille letters, while using 20 times less power than a standard processor.
Building on this work, the NUS team further enhanced robotic perception capabilities by combining vision and touch data in the spiking neural network. To do this, they commissioned a robot to classify several opaque containers containing different amounts of liquid, using sensory data from the artificial skin and a camera adapted to the context. In doing so, the researchers used the same vision and touch sensors to test the perception system’s ability to identify rotational slip, which is important for a stable grip.
The results
Once this sensory data was collected, the team sent it to a GPU and Intel’s Loihi neuromorphic research chip to compare processing capabilities. The results, which were presented at the ‘Robotics: Science and Systems’ conference this week, show that the combination of context-adapted vision and touch using a neural network of spikes allowed 10% more precision in object classification. compared to a vision-only system. In addition, they demonstrated the commitment of neuromorphic technology to power such robotic devices, and with Loihi they were able to process sensory data 21% faster than a high-performance GPU, while using 45 times less power.
