Research areas

Research areas

Many animals have versatile and robust sensory skills or useful and interesting behavior that researches would like to be able to exploit in robotic systems.

"Doing biology with robots"

At SDU Biorobotics, we conduct research in bio-inspired robots and in biorobotics. Bio-inspired robotics focuses on transferring good ideas from Nature to robotics. Here, the goal is to replicate the desired properties or skills of animals using engineering methods. Basic research in biorobotics, on the other hand, investigates the mechanisms that enable animals to perform as they do, by building working models of the mechanisms that may be responsible for their abilities. These models are normally robots, and the fundamental thrust of the research can be described as "doing biology with robots"

Neuromorphic Robots

We study the principles of the brain and how they can be used to design more intelligent and capable robots.

Artificial neural networks are being used extensively in solutions ranging from robot control  and machine learning to image recognition and game playing. Although the results are very impressive, they actually build on a very simplified model of biological neurons. Neuroscience has provided much more accurate models, but they are currently too complicated to directly implement in computers. Instead, scientists and industry are working on alternative computer architectures to support more brain-like computation.

Our work is centered around understanding the principles of the brain and applying those to control real robots. We work with Intel on applying their experimental computer chip Loihi as a drone controller, with SDU Vikings on applying new sensor technology to self-driving cars and with research groups all over the world on new sensors, computational models, and much more. We investigate novel approaches to learning based not only on synaptic adjustments, but also dopaminergic signals and adaptation of the neural structure. We test our hypotheses on real robots, be it walking, driving or flying, because we believe that having a body is prerequisite for intelligence.

World Economic Forum has named neuromorphic computing one of the top emerging technologies, promising solutions to many of the currently unsolvable problems and "to drive the next stage in miniaturization and artificial intelligence".  We do what we can every day to deliver on those expectations.

Embodied Neurorobotics

Living creatures, like walking animals, impress the observer by the elegance of their locomotion. They can also continuously learn and quickly adapt to new situations or solve specific tasks. They can even exploit their morphology to perform diverse complex autonomous behaviors, such as locomotion, object manipulation and transportation, and navigation, with a high degree of energy efficiency.

To achieve such complex behaviors for artificial walking systems, we focus on the Embodied Neurorobotics approach [1]. The approach not only (i) considers an integration of neural systems, body, and often the environment too, but also (ii) focuses on the detailed dynamical interactions of the neural computation (i.e., activity, plasticity, and memory at different time scales [2-4]) with the environment to realize and generalize complex autonomous behaviors. By doing so, we aim to understand the complex dynamical interactions between physical and computational components in embodied neural closed-loop systems.

Finally, we transfer our knowledge and development to other systems, such as service robots and exoskeletons, as well as to other application areas, such as service, construction, inspection, transportation, search and rescue, planetary exploration, and agriculture.

Central elements of embodied neurorobotics


[1] Manoonpong, P. and Tetzlaff, C. (2018) Editorial: Neural Computation in Embodied Closed-Loop Systems for the Generation of Complex Behavior: From Biology to Technology. Front. Neurorobot. 12:53. doi: 10.3389/fnbot.2018.00053

[2] Thor, M. and Manoonpong, P. (2019) Error-based Learning Mechanism for Fast Online Adaptation in Robot Motor Control, IEEE Transactions on Neural Networks and Learning Systems , doi: 10.1109/TNNLS.2019.2927737

[3] Aoi, S., Manoonpong, P., Ambe, Y., Matsuno, F. and Wörgötter, F. (2017) Adaptive Control Strategies for Interlimb Coordination in Legged Robots: A Review. Front. Neurorobot. 11:39. doi: 10.3389/fnbot.2017.00039

[4] Pitchai, M., Xiong, X., Thor, M., Billeschou, P., Lukas Mailander, P., Leung, B., Kulvicius, T., Manoonpong, P. (2019) CPG Driven RBF Network Control with Reinforcement Learning for Gait Optimization of a Dung Beetle-Like Robot. In: Tetko I., Kůrková V., Karpov P., Theis F. (eds) Artificial Neural Networks and Machine Learning – ICANN 2019: Theoretical Neural Computation. ICANN 2019. Lecture Notes in Computer Science, vol 11727. Springer, Cham

MOdular Robot Framework - MORF

The most frequently means to achieve mobility in robots is wheels, as their simplicity and low power consumption are appealing. However, wheeled robots are predominantly limited to flat surfaces with few obstacles and in need of additional mechanisms if manipulateon tasks are to be performed. This means the world often must change to fit the robot and not the other way around.

An alternative solution is legs. A legged robot needs legs with at least two degrees of freedom to move - one for lifting and one for swinging - but it is usually equipped with legs that have three degrees of freedom to allow additional manoeuvring. This increases power consumption and requires a more complex controller due to the complex body structure [1]. So why use legged robots at all? Legged robots are, first of all, able to interact with generic physical environments that are either designed for legged locomotion (humans) or rough terrain found in Nature [1].

Current solutions to adaptive locomotion for legged robots are promising, but often look miserable and are far from able to compete with the behaviors of real animals. This is presumably because the benefits of using legs most often are overshadowed by their high design complexity.

We, therefore, developed MORF, a MOdular Robot Framework which can be used in a wide range of studies. The primary aim of MORF is for it to be easy and convenient to use, such that researchers can focus more on the actual controller of the robot and not the hardware. Its design makes use of bio-inspired kinematics and state-of-the-art components to achieve high performance. This enables some of the complexity to be moved from the controller to the mechanics of the system.

MORF is modular as it defines standards that can be used for re-configuring, extending, and/or replacing parts of the robot, e.g. body shape. MORF can, for example, be configured either as an insect or a mammal. MORF furthermore includes a software suite with a full simulation of the robot.


[1] Todd, D. J. (1985). Walking Machines (Springer US). doi:10.1007/978-1-4684-6858-8_2 Kajita, S. and Espiau, B. (2008). Legged robots. In Springer Handbook of Robotics. 361–389. doi:10.1007/978-3-540-30301-5_17

For more information, please contact Mathias Thor

ESRL news

Human Robotics

Humans have remarkable sensorimotor learning capabilities. Our research integrates computational means and robotic platforms to investigate the principles underlying humans’ ability to quickly learn complex motor tasks. More specifically, we focus on the following questions: (i) How can humans learn neuromotor primitives to carry out complex tasks? (ii) How does learning explore and enable adaptation to individual biomechanics properties (e.g., mechanical impedance)? To answer these questions, experimental tools, such as EMGs, and software, such as OpenSim, are utilized to collect and analyze data from humans when performing tasks. By extracting the analytic principles, the computational models (e.g., sensorimotor learning and neuromechanics control) are developed and tested both in simulated and physical bio-robotic platforms, such as human-like robot arms. The robot platforms and the controllers in turn serve as inspiring means to stimulate new experiments with human motor control tasks. Taken together, human robotics paves a forward way to close the human-robot research loop. The results can contribute to the next generation exoskeleton augmentation technology, e.g., lightweight designs, user-friendly control, and seamless collaboration.
The cycle of human robotics