We observe that the self-organization of many simple and deformable elements creates a rich variety of physical adaptation in nature in a bottom-up manner. These examples exist on multiple scale lengths from embryonic cells to fish schools. Conversely, we follow a top-down approach to build robotic mechanisms to realize predefined designs that are optimized for certain tasks and environments.
I believe that the novel physically adaptive robots exist in the space that combines these two complementary design approaches. While soft deformable materials and self-assembling systems would allow us to explore unprecedented robotic forms and functions in a bottom-up manner, top-down control and machine learning methods would achieve repeatability and programmability of these novel systems. My research focuses on developing the tools and platforms to investigate this search space.
My research philosophy for achieving robotic physical adaptation.
A – Bioinspired Soft Robots
One of my earliest soft robots driven with bowden cables.
The depiction of a robot used to be not much different than an aluminum toaster for a long time. Inspirations from biology changed this outlook and introduced soft and deformable materials to robot designs. With the advances in materials research, now we can easily fabricate and functionalize such materials and build robots that demonstrate tasks that were not possible with conventional industrial robots. I always look at biology to find inspiration for the morphological design and functional variety for my robots. I benefit from the viscoelastic deformation of soft materials to build robots that demonstrate examples of bio-inspired sensing, locomotion, and manipulation.
Manipulation (not to be confused with the synonym term in psychology) is one of the fundamental functions for survival and adaptation in nature. From the most simple to the most complex, organisms have evolved certain body appendages to perform manipulation tasks that define their physical adaptation capabilities. It is no surprise that the understanding and analysis of this function has become one of the basic elements of robotics science, and my research. Taking the human hand as the template, I am specifically interested in the design and development of robotic manipulators that are both dexterous and physically compliant.
Our view on the balanced manipulator design.
The common principles of physically adaptive manipulators establish a balance between damping, forces, precision, and compliance.
Our anthropomorphic fingers flipping a coin.
The hybrid use of hard and soft materials allows the robotic manipulators to perform anthropomorphic tasks.
Just like manipulation, sensing is crucial for physical adaptation as it provides vital information about the changes that are constantly happening outside and inside an organism, and naturally a robot. My research focuses on the soft robotics sensing perspective, where I address the challenges in gathering tactile information from continuous and deformable robot bodies, and finding clever solutions to correlate visual information to soft robot kinematics.
Our robotic arm can fabricate sensors for different stimuli.
The in situ fabrication of soft sensors makes it possible for the robots to actively sense the stimuli in their environment.
Our soft sensor (the thin black line) can be applied directly onto the surface.
A functional conductive thermoplastic material allows the tactile sensing of dynamic deformations on a soft continuum surface.
Organisms and robots need to move from one place to another for different reasons, but the role of locomotion in physical adaptation is the same for both kingdoms. From the soft robotics perspective, the soft body elements used for locomotion may enable creative means of motion, but the control of the whole process becomes challenging compared to conventional rigid body dynamics systems. My research focuses on designing the compliant robotic locomotion appendages and exploring their efficient and adaptive control architectures.
Our robot can descend in free space using its dragline like a spider.
Inspired from spiders, the in situ fabrication of a dragline with soft materials allows free space locomotion.
We integrated an actuated flexible spine to increase bounding efficiency.
The compliant musculoskeletal system enables efficient and adaptive terrestrial locomotion both for animals and robots.
- Myorobotics project in the International Symposium on Adaptive Motion of Animals and Machines (2013),
- International Conference on Robotics and Automation (2011).
B – Self-Assembly
Self-assembly is a ubiquitous process in both the living and non-living systems that generate complex and functional structures from the local interactions between a large set of simpler components. Especially in biology, physical adaptation is explained by linking the emergence of functions to the formation of such structures as a result of the collective behavior of neighboring cells.
The ability to control the self-assembly process is very exciting for the robotics field as it may enable alternative design methods to build novel robotic systems. This is also an interdisciplinary line of research that will additionally elucidate fundamental physics and enable the fabrication of novel functional materials.
My research focuses on the design of physical mechanisms that allow the programming of the self-assembly pathway of robotic systems.
- Proceedings of the National Academy of Sciences (2020).
- Guided self-assembly of soft robots, coming soon.
C – Machine Learning
Soft robots have great potential in achieving physical adaptation via viscoelastic deformations. However, the features that enable this potential (e.g., virtually infinite degrees of freedom, material-enabled behavior) make it challenging to model and predict robot kinematics and dynamics, and apply conventional control methods. Also, common approximation methods depending on body discretization and constant curvature assumptions fall short of revealing the true potential of these robots.
I believe that state-of-the-art machine learning methods may help us explore the vast design and function space of soft robots. With this approach, we may not only unveil unprecedented robotic systems but also learn their inherent complex models of design and control.
My current work focuses on applying suitable machine learning algorithms to learn the soft robotic function and design parameters, and developing the platforms that allow repeatable and reliable machine learning experiments on physical systems.
- Robotics: Science and Systems (2020),
- Task space adaptation, coming soon.