MOTIF Brings Human-Like Sensing to Robots
USC engineers built MOTIF, a sensor-packed robotic hand that mimics human touch by merging thermal, visual, depth, and tactile inputs.
It is hard to beat the human hand when it comes to sensing and interacting with the physical world. So hard, in fact, that the best artificial systems dreamed up by today’s top robotics engineers do not even come close. The complex network of sensory receptors in the skin that, in concert with nerve fibers and the brain, recognize pressure, vibration, texture, temperature, and pain still hold many mysteries that we have not yet uncovered, let alone reproduced.
That leaves roboticists in a tough spot. Without the incredible sensitivity of human touch, it is all but impossible to design a robotic hand that can interact with the physical world in a human-like way. Getting to that next level of performance will undoubtedly take at least a few significant technological breakthroughs. But in the meantime, a group of engineers at the University of Southern California has introduced a stopgap measure that might help us to design machines that are at least nearer to the ultimate goal.
The team has designed a robotic hand named MOTIF (Multimodal Observation with Thermal, Inertial, and Force sensors) that takes something of a kitchen sink approach to the problem. It may be the case that none of our sensing systems are as sensitive as those found in the human hand, but if we throw a wide variety of sensors at the problem, we might be able to stitch the data together to produce a clearer picture than we had before. To test out that hypothesis, the researchers included everything from depth sensors, thermal cameras, inertial measurement units, and a camera into MOTIF.
MOTIF is built on top of the LEAP hand, a well-known open-source dexterous robotic hand platform. The team then loaded this existing hand down with sensors. It includes thin film tactile sensors covering each finger, allowing for dense, high-resolution tactile data collection. These sensors provide a 6x6 sampling grid with 2.5 mm resolution, enabling the hand to detect force magnitude with a trigger threshold of just 20 grams.
To capture spatial data, the hand includes a time-of-flight depth sensor embedded in the palm for real-time distance measurement. A FLIR Lepton 3.5 thermal camera provides infrared temperature data, initially collected at a resolution of 160x120 pixels that is scaled up to 1280×960 via interpolation. This enables the hand to detect heat patterns across surfaces and guide its grasp accordingly.
Each of the hand’s finger joints is equipped with a motion tracking unit combining a 3-axis gyroscope and 3-axis accelerometer, with a 3-axis magnetoresistive sensor added for nine-axis orientation tracking. These sensors are shielded from electromagnetic interference using conductive tape, ensuring clean signals even in the proximity of servomotors.
On the wrist, a Raspberry Pi Camera Module 2 captures high-resolution RGB visual data at 30 frames per second, giving the hand environmental context and object recognition capabilities. A Raspberry Pi 5 serves as the primary data processor for the entire system, handling the integration of all of the sensory inputs.
The MOTIF hand is open source and designed to be reproducible for under $4,000. With continued development, including the possible addition of high-resolution fingertip sensors like the Digit 360, the MOTIF hand could soon serve as an important platform for advancing robotic manipulation in both research and real-world applications.