Robotics and Biology Laboratory

RBO Mission Statement

Robotics and Biology Laboratory

How can we create robots that are truly intelligent and autonomous, not just in the context of some specific and narrow setting but in some broad and general sense? This requires building a truly intelligent embodied system.  It is a formidable challenge both in terms of engineering and science. Autonomous systems capable of intelligent behavior must tightly integrate many capabilities in synergistic ways, challenging our ability to build truly robust systems. At the same time, we still lack a foundational scientific understanding of how to generate versatile, robust, adaptable behavior in general, everyday environments.

Research in the Robotics and Biology Laboratory approaches these challenges in the context of real-world problems, including manipulation, in-hand manipulation, interactive perception, learning from demonstration, co-design of manipulation skills, and system building. We closely collaborate with researchers from psychology and behavioral biology to support them in their quest to understand the behavior of biological agents and in turn learn about these agents and the methods employed. Imported into robotics, both methods and understanding push our ability to build better robots. Still, we are far away from matching human abilities but we have begun laying a conceptual foundation that one day might.

Our approach to the aforementioned challenges in science and engineering is inspired by insights from biological intelligent agents. But we also take advantage of the achievements of synthetic disciplines, of course. We work in a problem-centric manner, not by pushing a particular method.  Instead, we employ whatever method appears most suitable to the problem. We identify general principles and explore their power for intelligent agency. So far, this has been a journey across disciplinary boundaries generally accepted but seldomly questioned. Instead of using general-purpose, generic robotic platforms, we believe that embodiment and control must be considered together; this is generally called co-design. Instead of attributing the responsibility of intelligent behavior exclusively to the robot (or even the "brain") itself, we believe that robustness and generality arises from the interplay of a robot's software and hardware with its environment. As Rod Brooks famously said: "The world is its own best model." And the world can serve in many other ways to augment the ability of machines. We have shown that by considering perception and manipulation as two sides of a single coin, we can produce very robust and general robot behavior; we have termed this Interactive Perception. These are just a few of the insights we believe will ultimately contribute to the foundational principles of generating intelligent behavior. And, after all, that is the goal of robotics.

© RBO

Protein Structure Prediction

Proteins are one of the most abundant molecules in living organisms. They are in charge of a variety of crucial functions, such as transporting molecules (e.g. Hemoglobin), catalyzing reactions (e.g. Enzymes), replicating DNA, identifying and neutralizing foreign bacteria and viruses (e.g. Antibodies) and many more. Predicting the protein structures can help us understand how they function, to create new drugs targeting them, to understand how mutation affect them.

Interactive Perception © RBO

Interactive Perception

Interactive Perception is about acting to improve perception. The fundamental assumption is that the domains of perception and action cannot be separated, but form a complex which needs to be studied in its entirety. Using this approach, we try to design robot that explore their environment actively, in a way that reminds of how a baby explores a new toy.

© RBO

Manipulation Planning

Even though grasping has been a central topic in robotics for decades, robots still have great difficulty to pick up arbitrary objects when they operate in open, unknown environments or under uncontrolled conditions. Lifting this limitation would make using robots fit for many repetitive chores: robots could check and restock supermarket aisles, tidy up households, dispatch mail orders at distribution centers, collect ripe fruits, or do ecological pest control by selectively removing bugs.

© RBO

Sensorization of Soft Hands and Actuators

Soft robots are highly compliant due to the flexible materials they are made of. To observe and react to the robot's interaction with its environment, we need sensors. But because the robots are very soft, we can no longer use traditional, "hard" sensors. Instead, we develop novel, "soft" sensor technologies and research ways of combining sensor data with clever computation to extract relevant measurements from our soft robots.

In-Hand Manipulation

We design and investigate promising new control schemes in the quest for humanlike dexterity. Embodied Control is both computational and morphological; we use the soft anthropomorphic RBO Hand 3, which affords rich physical interactions and challenges us to think of unconventional algorithms.

© RBO

Representation Learning

Truly versatile robots cannot rely on representations that are specifically hand-crafted for every task. Instead they must be able to learn these representations from experience.

© RBO

Human Grasping

We study human grasping under a variety of conditions in order to identify and characterize different grasping strategies. Specially, we are interested in strategies that are robust to be performed under different kinds of impairment, e.g. visual. In addition to subjects' grasping with their natural hands, we also observe them when they use soft robotic hands such as RBO Hand 2 and Pisa/IIT Hand. Our final goal, is to transfer those robust strategies to a robot.

© RBO

Soft Hands

Soft Hands represent a departure from classical robot hand design, which often relies on exact models and precise planning of contact points. Instead, we aim to increase robustness and safety through the use of soft materials and flexible mechanics. This softness allows us to exploit contact with the environment and use it in robust grasping and manipulation strategies.

We developed the RBO Hand 3 to understand the necessary soft-robotic aspects, and formulate the concept of morphological-computation.

© RBO

Protein Motion

Proteins are the building blocks of life, participating in almost all cellular processes in our body. They transport smaller molecules, catalyze reactions, regulate the metabolism or support our immune system. All these functions require intensive interaction with other molecules, which is only possible if the shape of the interacting partners fit together. To allow such a fit, a protein continuously has to change and adjust its three-dimensional shape (Fig. 1). Understanding the motion of a protein, therefore, provides insights into the function it can perform. Most importantly, it can enable therapeutical intervention in order to regulate this function, resulting in drugs targeting severe diseases such as Alzheimer's, AIDS or cancer.