The mission of MERGe Lab is to build better robots through intentional design of their bodies. As robots are deployed outside of the lab and the factory, they will encounter a wide range of human environments and contexts. We focus our core research on sensor and actuator design, as these are the main ways that robots interact and perceive their environment.
Primary students: David B., Emiliano, Zach
The materials that a robot’s body is made of has significant downstream effects on the robot’s overall mechanical behavior and end performance. Thus, we are investigating how modifying the mechanical properties of a robot’s materials creates more effective robot designs. Our lab is pursuing two approaches: (1) multi-modulus 3D printing to print robots with continuous stiffness gradients, in collaboration with the ZAP Research Group and (2) using geometry as a design variable by making robots out of mechanical metamaterials / architected materials. We are particularly interested in fusing these two approaches into one coherent design philosophy, as it is currently an open question of which properties are better achieved through material vs. geometric formulations.
Primary students: Bill, Andrew, Tanish
Since many of our core sensors and actuators are based off of 3D printing, we have significantly more design control over a structure’s geometry, including sensor placement and layout. While we have demonstrated many promising examples of manually-designed fluidically innervated structures, leveraging computational methods and simulation offers enormous potential to create optimized robots. We are currently building parameterized models of our metamaterial actuators, as well as soft body simulations of our fluidically innervated sensors, enabling us to explore automatic routing and layout of our structures.
Primary students: Siqi, Benito, Joseph, Manav
Although current robots can do impressive grasping tasks with rigid grippers and computer vision, they still lack the contact-rich dexterity seen in human manipulation. We are working to create the hardware platforms that will enable systematic investigation needed to achieve complex manipulation. This ranges from incorporating high-resolution tactile sensors into grippers to creating multi-fingered robotic platforms that allow direct comparison of different hand layouts. Our tactile sensorization efforts have led to significant advances in manipulation, including manipulating submerged objects across the air-water interface. Our sensors can accurately measure grasping forces from 0-8 N with an average error of 0.2 N with a sample rate of 2 kHz, enabling perception of fast slip events (within 100 ms).
Primary students: David G., Aryaman, Caroline, Aileen
As robots increasingly move towards humanoid form factors, wearable devices that can measure human movement become increasingly more important, whether for data capture or for clinical applications. Sensors that can pick up desired signals across the diverse range of human(oid) morphologies are particularly challenging to create, especially given portability and interface concerns. We are building force sensors for humans and robots alike, ranging from shoes to protective sports equipment to lower-limb exoskeletons. We are actively collaborating with dancers and kinesiologists to better match the performance needs of expert human movements.
We are always on the lookout for new fields and techniques to merge into our research approach. Some current student-led directions include:
Some areas of interest for the PI include:
Our systems have outperformed similar soft robots in power efficiency (20x more efficient) and speed (2x faster), outperformed similar modular robots in locomotion (10x faster) while maintaining a high strength-weight ratio (76x), and created the largest sensorized soft robotic dataset (18 hours).
We then connected multiple auxbots together for more complex actuation, similar to how biological cells come together to form tissues and organs. With the speed and size optimized auxbots, we created a 2x2x2 cube that could move forward, turn, and flip, due to the high impulse response. Meanwhile, with the force optimized auxbots, we created a flipper-like locomotion by tying adjacent auxbots together with a wire. This simultaneous lifting and bending action led to a seven-bot quadruped that could carry loads up to 1.5x its total weight (2 kg), even with some auxbots stalling out. Auxbots thus demonstrate not only how geometry gives greater control of robot performance but also offers a potential pathway to address the gap between simulation and reality.
This approach enabled us to create several soft robots just by changing the geometric structure. We created a 4 degree-of-freedom robotic platform directly from HSAs by placing opposite handed HSAs in a 2x2 grid and attaching a motor to each HSA. Similar high degree-of-freedom platforms would require significantly more infrastructure, like the six prismatic joints needed for a Stewart platform. Likewise, we created robotic fingers from the HSAs by drawing a line through the pattern. This line would act as a strain-limiting layer, forcing the entire structure to bend inwards. Since the HSA gripper is motor-driven, it is 2x faster at opening and closing, 20x more power efficient, and occupies a smaller footprint than standard soft pneumatic-based grippers, all while maintaining a similar grasping performance.