Imagine a continuous soft robotic arm flexing around a grape or broccoli, adjusting its grip in real time because it lifts the article. Unlike traditional rigid robots, which generally aim to avoid contact with the environment as much as possible and avoid people for safety reasons, this arm senses subtle forces and stretches and bends in a way that more closely mimics the flexibleness of a human hand. Every movement is calculated to avoid excessive effort and complete the duty efficiently. In the laboratories of the MIT Computer Science and Artificial Intelligence Laboratory (CSAIL) and the Laboratory for Information and Decisions Systems (LIDS), these seemingly easy movements are the culmination of complex mathematics, careful engineering, and a vision for robots that may safely interact with people and delicate objects.
Soft robots, with their deformable bodies, promise a future wherein machines move more seamlessly alongside humans, assisting with care or handling delicate objects in industrial environments. But it’s precisely this flexibility that makes it difficult to regulate. Small bends or twists can create unpredictable forces and increase the chance of harm or injury. This motivates the necessity for secure control strategies for soft robots.
“Inspired by advances in secure control and formal methods for rigid robots, we aim to use these ideas to soft robotics—by modeling their complex behavior and incorporating contact somewhat than avoiding it—to enable more powerful designs (e.g., larger payload and precision) without sacrificing safety or embodied intelligence,” says senior lead creator and MIT Assistant Professor Gioele Zardini, the principal investigator at LIDS and the Department of Civil Engineering Environmental Engineering and is an affiliated faculty of the Institute for Data, Systems and Society (IDSS). “This vision is shared by current and parallel work by other groups.”
Safety comes first
The team developed a brand new framework that mixes nonlinear control theory (control of systems with highly complex dynamics) with advanced physical modeling techniques and efficient real-time optimization to realize what they call “contact-aware safety.” The approach focuses on high-order control barrier functions (HOCBFs) and high-order control barrier functions (HOCLFs). HOCBFs define secure operating limits and be sure that the robot doesn’t exert unsafe forces. HOCLFs efficiently guide the robot to its task goals while ensuring a balance between safety and performance.
“Essentially, we’re teaching the robot to know its own limits when interacting with the environment while achieving its goals,” says Kiwan Wong, a graduate student within the MIT Department of Mechanical Engineering and lead creator of a brand new paper describing the framework. “The approach involves a fancy derivation of soppy robot dynamics, contact models and control constraints, however the specification of control objectives and safety barriers is fairly easy for the practitioner and the outcomes are very tangible as one sees the robot moving easily, responding to contact and never causing unsafe situations.”
“Compared to traditional kinematic CBFs – where forward-invariant secure quantities are difficult to specify – the HOCBF framework simplifies barrier design and its optimization formulation takes into consideration system dynamics (e.g. inertia) and ensures that the soft robot stops early enough to avoid unsafe contact forces,” says Wei Xiao, assistant professor at Worcester Polytechnic Institute and former CSAIL postdoctoral fellow.
“Since the arrival of soppy robots, the sphere has highlighted their embodied intelligence and greater inherent safety in comparison with rigid robots, because of passive material and structural compliance. But their 'cognitive' intelligence – particularly safety systems – lags behind that of rigid serial-connected manipulators,” says co-lead creator Maximilian Stölzle, research intern at Disney Research and former PhD student at Delft University of Technology and visiting researcher at MIT LIDS and CSAIL. “This work helps close this gap by adapting proven algorithms to soft robots and tailoring them for secure contact and soft continuum dynamics.”
The LIDS and CSAIL team tested the system in a series of experiments geared toward testing the robot's safety and adaptableness. In one test, the arm pressed gently against a compliant surface while maintaining precise force without overshooting. In one other case, it traced the contours of a curved object and adjusted its grip to forestall it from slipping. In one other demonstration, the robot manipulated fragile objects alongside a human operator, responding in real time to unexpected shocks or movements. “These experiments show that our framework is capable of generalize to different tasks and goals, and that the robot can recognize, adapt and act in complex scenarios while at all times respecting clearly defined safety boundaries,” says Zardini.
Of course, soft robots with contact-sensitive security could add real value in places where the stakes are high. In healthcare, they might assist in surgeries by enabling precise manipulations while reducing risk to patients. In industry, they’ll handle fragile goods without constant supervision. In home environments, robots could help with household chores or care tasks and safely interact with children or the elderly – a crucial step in making soft robots reliable partners in real-world environments.
“Soft robots have incredible potential,” says co-lead creator Daniela Rus, director of CSAIL and professor within the Department of Electrical Engineering and Computer Science. “But ensuring safety and encoding movement tasks over relatively easy targets has at all times been a key challenge. We desired to create a system wherein the robot can remain flexible and responsive while mathematically guaranteeing that it doesn’t exceed secure force limits.”
Combination of soppy robot models, differentiable simulation and control theory
The control strategy relies on a differentiable implementation of the so-called Piecewise Cosserat Segment (PCS) dynamics model, which predicts how a soft robot deforms and where forces accumulate. This model allows the system to predict how the robot's body will reply to operations and complicated interactions with the environment. “The aspect I like most about this work is the combination of integrating latest and old tools from different fields equivalent to advanced soft robot models, differentiable simulation, Lyapunov theory, convex optimization and injury severity-based safety constraints. All of that is well integrated right into a real-time controller based entirely on first principles,” says co-author Cosimo Della Santina, associate professor at Delft University of Technology.
This is complemented by the Differentiable Conservative Separating Axis Theorem (DCSAT), which estimates distances between the soft robot and obstacles within the environment, which may be approximated in a differentiable manner using a series of convex polygons. “Previous differentiable distance metrics for convex polygons either didn’t calculate penetration depth – essential for estimating contact forces – or provided non-conservative estimates that would compromise safety,” says Wong. “Instead, the DCSAT metric provides strictly conservative and subsequently secure estimates while allowing for fast and differentiable computation.” Together, PCS and DCSAT give the robot a predictive sense of its environment for more proactive and safer interactions.
Looking forward, the team plans to increase their methods to three-dimensional soft robots and explore integration with learning-based strategies. By combining contact-aware safety with adaptive learning, soft robots could handle much more complex, unpredictable environments.
“That’s what makes our work exciting,” says Rus. “You can see that the robot behaves human-like and cautious, but behind this grace lies a strict framework of control that ensures that it never exceeds its limits.”
“Soft robots are generally safer to interact with than rigid robots due to the compliance and energy-absorbing properties of their bodies,” says Daniel Bruder, an assistant professor on the University of Michigan who was not involved within the research. “However, as soft robots turn out to be faster, stronger and more powerful, this may increasingly now not be enough to make sure safety. This work represents a critical step in ensuring that soft robots can operate safely by providing a technique for limiting contact forces across their entire bodies.”
The team's work was supported partially by grants from the Hong Kong Jockey Club, the European Union's Horizon Europe program, the Cultuurfonds Wetenschapsbeurzen, and the Rudge (1948) and Nancy Allen Chair. Their work was published earlier this month within the Institute of Electrical and Electronics Engineers.
