Publications

2021

1. Versatile modular neural locomotion control with fast learning Mathias Thor, and Poramate Manoonpong Preprint arXiv: 2107.07844 2021 Abstract [HTML] [PDF]   Preprint paper

   Legged robots have significant potential to operate in highly unstructured environments. The design of locomotion control is, however, still challenging. Currently, controllers must be either manually designed for specific robots and tasks, or automatically designed via machine learning methods that require long training times and yield large opaque controllers. Drawing inspiration from animal locomotion, we propose a simple yet versatile modular neural control structure with fast learning. The key advantages of our approach are that behavior-specific control modules can be added incrementally to obtain increasingly complex emergent locomotion behaviors, and that neural connections interfacing with existing modules can be quickly and automatically learned. In a series of experiments, we show how eight modules can be quickly learned and added to a base control module to obtain emergent adaptive behaviors allowing a hexapod robot to navigate in complex environments. We also show that modules can be added and removed during operation without affecting the functionality of the remaining controller. Finally, the control approach was successfully demonstrated on a physical hexapod robot. Taken together, our study reveals a significant step towards fast automatic design of versatile neural locomotion control for complex robotic systems.

2020

1. Adaptive Neural Control for Efficient Rhythmic Movement Generation and Online Frequency Adaptation of a Compliant Robot Arm Florentijn Degroote, Mathias Thor, Jevgeni Ignasov, Jørgen Christian Larsen, Emilia Motoasca, and Poramate Manoonpong In the 27th International Conference on Neural Information Processing 2020 Abstract [HTML]   Conference paper

   In this paper, we propose an adaptive and simple neural control approach for a robot arm with soft/compliant materials, called GummiArm. The control approach is based on a minimal two-neuron oscillator network (acting as a central pattern generator) and an error-based dual integral learning (DIL) method for efficient rhythmic movement generation and frequency adaptation, respectively. By using this approach, we can precisely generate rhythmic motion for GummiArm and allow it to quickly adapt its motion to handle physical and environmental changes as well as interacting with a human safely. Experimental results for GummiArm in different scenarios (e.g., dealing with different joint stiffnesses, working against elastic loads, and interacting with a human) are provided to illustrate the effectiveness of the proposed adaptive neural control approach.

2. iCrawl: An Inchworm-Inspired Crawling Robot M. B. Khan, T. Chuthong, C. Danh Do, M. Thor, P. Billeschou, J. C. Larsen, and P. Manoonpong IEEE Access 2020 Abstract [HTML]   Journal paper

   Inchworms use their morphology and evolved behaviors to crawl and climb various complex surfaces. This has inspired the development of different robots that can demonstrate similar capabilities for various applications such as the inspection of a complex environment. One of the key challenges in designing these robots is to enable them to be practically deployable with a compact design for providing continuous adaptability to a complex terrain such as an outer-pipe surface. Taking this into consideration, we present a new design for an inchworm-inspired crawling robot (iCrawl). The 5 DOF robot relies on two legs; each with an electromagnetic foot, in order to crawl on the metal pipe surfaces. The robot uses a passive foot-cap underneath an electromagnetic foot, enabling it to be a versatile pipe-crawler. The foot-cap design is an abstraction of the leg posture in an inchworm adapting to a round surface. The proposed foot-caps give the robot adaptability and stability for crawling on metal pipes of various curvatures. A state-machine based controller was developed to produce the required motor signals for the two inchworm-inspired crawling gaits: i) the step gait, and ii) the sliding gait. Both gaits were tested on the robot, eventually leading to it effectively crawling on the pipes and flat surfaces, climbing a metal wall and a pipe, and succeeding in obstacle avoidance during crawling. Experimental results also show that the robot has the ability to crawl on the metal pipes of various curvatures using the foot-caps and an appropriate gait. The robot can be used as a new robotic solution to assist close inspection outside the pipelines, thus minimizing downtime in the oil and gas industry.

3. Generic Neural Locomotion Control Framework for Legged Robots M. Thor, T. Kulvicius, and P. Manoonpong IEEE Transactions on Neural Networks and Learning Systems 2020 Abstract [HTML] [PDF]   Journal paper

   In this paper, we present a generic locomotion control framework for legged robots and a strategy for control policy optimization. The framework is based on neural control and black-box optimization. The neural control combines a central pattern generator (CPG) and a radial basis function (RBF) network to create a CPG-RBF network. The control network acts as a neural basis to produce arbitrary rhythmic trajectories for the joints of robots. The main features of the CPG-RBF network are: 1) it is generic, since it can be applied to legged robots with different morphologies; 2) it has few control parameters, resulting in fast learning; 3) it is scalable, both in terms of policy/trajectory complexity and the number of legs that can be controlled using similar trajectories; 4) it does not rely heavily on sensory feedback to generate locomotion and is thus less prone to sensory faults; and 5) once trained, it is simple, minimal, and intuitive to use and analyze. These features will lead to an easy-to-use framework with fast convergence and the ability to encode complex locomotion control policies. In this work, we show that the framework can successfully be applied to three different simulated legged robots with varying morphologies, and even broken joints, to learn locomotion control policies. We also show that after learning, the control policies can also be successfully transferred to a real-world robot without any modifications. We, furthermore, show the scalability of the framework by implementing it as a central controller for all legs of a robot and as a decentralized controller for individual legs and leg pairs. By investigating the correlation between robot morphology and encoding type, we are able to present a strategy for control policy optimization. Finally, we show how sensory feedback can be integrated into the CPG-RBF network to enable online adaptation.

2019

1. CPG Driven RBF Network Control with Reinforcement Learning for Gait Optimization of a Dung Beetle-Like Robot Matheshwaran Pitchai, Xiaofeng Xiong, Mathias Thor, Peter Billeschou, Peter Lukas Mailänder, Binggwong Leung, Tomas Kulvicius, and Poramate Manoonpong In Artificial Neural Networks and Machine Learning – ICANN 2019: Theoretical Neural Computation 2019 Abstract [HTML]   Conference paper

   In this paper, we employ a central pattern generator (CPG) driven radial basis function network (RBFN) based controller to learn optimized locomotion for a complex dung beetle-like robot using reinforcement learning approach called “Policy Improvement with Path Integrals (PI²)”. Our CPG driven RBFN controller is inspired by rhythmic dynamic movement primitives (DMPs). The controller can be also seen as an extension to a traditional CPG controller, which usually controls only the frequency of the motor patterns but not the shape. Our controller uses the CPG to control the frequency while the RBFN takes care of the shape of the motor patterns. In this paper, we only focus on the shape of the motor patterns and optimize those with respect to walking speed and energy efficiency. As a result, the robot can travel faster and consume less power than using only the CPG controller.

2. Modular Neural Control for Dung Beetle-like Leg Movements of a Dung Beetle-like Robot Binggwong Leung, Peter Billeschou, Mathias Thor, and Poramate Manoonpong In The 9th International Symposium on Adaptive Motion of Animals and Machines (AMAM 2019) 2019 Abstract [HTML]   Conference paper

   Dung beetles can perform impressive multiple motor behaviors using their legs. The behaviors include walking and rolling a large dung ball on different terrains, e.g., level ground and different slopes. To achieve such complex behaviors for legged robots, we propose here a modular neural controller for dung beetle-like locomotion and object transportation behaviors of a dung beetle-like robot. The modular controller consists of several modules based on three generic neural modules. The main modules include 1) a neural oscillator network module (as a central pattern generator (CPG)), 2) a neural CPG postprocessing module (PCPG), 3) a velocity regulating network module (VRN). The CPG generates basic rhythmic patterns. The patterns are first shaped by the PCPG and their amplitudes as well as phases are later modified by the VRN to obtain proper motor patterns for locomotion and object transportation. Combining all these neural modules, we can achieve different motor patterns for four different actions which are forward walking, backward walking, levelground ball rolling, and sloped-ground ball rolling. All these actions can be activated by four input neurons. The experimental results show that the simulated dung beetle-like robot can robustly perform the actions. The average forward speed is 0.058 cm/s and the robot is able to roll a large ball (about 3 times of its body height and 2 times of its weight) up different slope angles up to 25 degrees.

3. A Fast Online Frequency Adaptation Mechanism for CPG-Based Robot Motion Control M. Thor, and P. Manoonpong IEEE Robotics and Automation Letters 2019 Abstract [HTML] [PDF]   Journal paper

   In this letter, we present an online learning mechanism called the dual integral learner for fast frequency adaptation in neural central pattern generator (CPG) based locomotion control of a hexapod robot. The mechanism works by modulating the CPG frequency through synaptic plasticity of the neural CPG network. The modulation is based on tracking error feedback between the CPG output and joint angle sensory feedback of the hexapod robot. As a result, the mechanism will always try to match the CPG frequency to the walking performance of the robot, thereby ensuring that the entire generated trajectory can be followed with low tracking error. Real robot experiments show that our mechanism can automatically generate a proper walking frequency for energy-efficient locomotion with respect to the robot body as well as being able to quickly adapt the frequency online within a few seconds to deal with external perturbations such as leg blocking and a variation in electrical power. These important features will allow a hexapod robot to be more robust and also extend its operating time. Finally, the generality of the mechanism is shown by successfully applying it to a compliant robotic manipulator arm called GummiArm.

4. Error-Based Learning Mechanism for Fast Online Adaptation in Robot Motor Control M. Thor, and P. Manoonpong IEEE Transactions on Neural Networks and Learning Systems 2019 Abstract [HTML] [PDF]   Journal paper

   Existing state-of-the-art frequency adaptation mechanisms of central pattern generators (CPGs) for robot locomotion control typically rely on correlation-based learning. They do not account for the tracking error that may occur between the actual system motion and CPG output, leading to the loss of precision, unwanted movement, inefficient energy locomotion, and in the worst cases, motor collapse. To overcome this problem, we developed online error-based learning for frequency adaptation of CPGs. The learning mechanism used for error reduction is a novel modification of the dual learner (DL) called dual integral learner (DIL). Being able to reduce tracking and steady-state errors, it can also perform fast and stable learning, adapting the CPG frequency to match the performance of robotic systems. Control parameters of the DIL are more straightforward for complex systems (like walking robots), compared to traditional correlation-based learning, since they correspond to error reduction. Due to its embedded memory, the DIL can relearn quickly and recover spontaneously from the previously learned parameters. All these features are not covered by the existing frequency adaptation mechanisms. We integrated the DIL into a neural CPG-based motor control system for use on different legged robots with various morphologies for evaluation. The results show that: 1) the DIL does not require precise adjustment of its parameters to fit specific robots; and 2) the DIL can automatically and quickly adapt the CPG frequency to the robots such that the entire trajectory of the CPG can be precisely followed with very low tracking and steady-state errors. Consequently, the robots can perform the desired movements with more energy-efficient locomotion compared to the state-of-the-art correlation-based learning mechanism called frequency adaptation through fast dynamical coupling (AFDC). In the future, the proposed error-based learning mechanism for fast online adaptation in robot motor control can be used as a basis for trajectory optimization, universal controllers, and other studies concerning the change of intrinsic or extrinsic parameters.

5. MORF - Modular Robot Framework University of Southern Denmark - Campusvej 55, 5230 Odense, Denmark 2019 [PDF]   Master Thesis   Top grade

2018

1. A dung beetle-inspired robotic model and its distributed sensor-driven control for walking and ball rolling M. Thor, T. Strøm-Hansen, L. B. Larsen, A. Kovalev, S. N. Gorb, E. Baird, and P. Manoonpong Artificial Life and Robotics 2018 Abstract [HTML]   Journal paper

   A typical approach when designing a bio-inspired robot is to simplify an animal model and to enhance the functionality of interest. For hexapod robots, this often leads to a need of supplementary mechanics to become multifunctional. However, a preferable solution is to employ the embodied multifunctional capabilities of the animal as inspiration for robot design. Using this approach, we present a method for translating the kinematic chain of a dung beetle from which an accurate kinematic model and a simplified one were simulated and compared. The beetle was selected due to its multifunctional locomotory capabilities including walking as well as standing on and rolling a ball. For testing the models, we developed a distributed sensor-driven controller that can generate walking and ball-rolling behaviors. A comparison of the two modeling approaches shows a similar performance with regards to walking stability and accuracy, but differences when it comes to speed and multifunctionality. This is because the accurate model is able to use its legs to walk faster and roll a ball, which the simplified one is not. In conclusion, the accurate model of a dung beetle-inspired robot is advantageous as it, together with our novel control mechanism, is able to elicit behaviors comparable to those of the real dung beetle (i.e., walking and rolling a dung ball).

2. Modular neural control for bio-inspired walking and ball rolling of a dung beetle-like robot Binggwong Leung, Mathias Thor, and Poramate Manoonpong In ALife 2018 2018 Abstract [HTML] [PDF]   Conference paper

   Dung beetles can perform impressive multiple motor behaviors using their legs. The behaviors include walking and rolling a large dung ball on different terrains, e.g., level ground and different slopes. To achieve such complex behaviors for legged robots, we propose here a modular neural controller for dung beetle-like locomotion and object transportation behaviors of a dung beetle-like robot. The modular controller consists of several modules based on three generic neural modules. The main modules include 1) a neural oscillator network module (as a central pattern generator (CPG)), 2) a neural CPG postprocessing module (PCPG), 3) a velocity regulating network module (VRN). The CPG generates basic rhythmic patterns. The patterns are first shaped by the PCPG and their amplitudes as well as phases are later modified by the VRN to obtain proper motor patterns for locomotion and object transportation. Combining all these neural modules, we can achieve different motor patterns for four different actions which are forward walking, backward walking, levelground ball rolling, and sloped-ground ball rolling. All these actions can be activated by four input neurons. The experimental results show that the simulated dung beetle-like robot can robustly perform the actions. The average forward speed is 0.058 cm/s and the robot is able to roll a large ball (about 3 times of its body height and 2 times of its weight) up different slope angles up to 25 degrees.

3. MORF - Modular Robot Framework Mathias Thor, Jørgen Christian Larsen, and Poramate Manoonpong In The 2nd International Youth Conference of Bionic Engineering (IYCBE 2018) 2018 [HTML] [PDF]   Conference paper   Best Student Paper Award

2017

1. Advantages of using a biologically plausible embodied kinematic model for enhancement of speed and multifunctionality of a walking robot Mathias Thor, Theis Strøm-Hansen, and Leon Bonde Larsen In Proceedings of the 2nd International Symposium on Swarm Behavior and Bio-Inspired Robotics (SWAM 2017) 2017 Abstract [PDF]   Conference paper   Best Student Paper Award

   A typical approach when designing a bio-inspired robot is to simplify an animal model and to enhance the functionality of interest. For hexapod robots, this often leads to models with an unknown performance; thereby influencing their functionality and requiring supplementary mechanics to become multifunctional. A preferable solution is to employ the embodied multifunctional capabilities of the animal as inspiration for robot design. For this reason, we present a method for translating the kinematic chain of a dung beetle, from which an accurate model and a simplified one were created and simulated. A comparison between the two modeling approaches shows a similar performance in regards to walking stability and accuracy, but differences when it comes to speed and multifunctionality. Here the accurate model outperforms the simplified model, by both walking faster and being capable of performing additional locomotory tasks. In conclusion, the accurate model of a dung beetle-inspired robot is advantageous in regards to multifunctional abilities including walking as well as standing on and rolling a ball, like a dung beetle.

2. Distributed Sensor-Driven Control for Bio-Inspired Walking and Ball Rolling of a Dung Beetle-Like Robot Theis Strøm-Hansen, Mathias Thor, Leon Bonde Larsen, Emily Baird, and Poramate Manoonpong In Proceedings of the 2nd International Symposium on Swarm Behavior and Bio-Inspired Robotics (SWAM 2017) 2017 Abstract [PDF]   Conference paper

   Bio-inspired robotics is an approach that looks at how nature solves a (complex) problem and applies the solution to a robot. Based on this approach, we have looked at walking animals that can exploit their legs for multiple functions, like locomotion and object manipulation/transportation. This is to expand the ability of walking robots. According to this, we have developed our robotic model based on a dung beetle, an animal that can walk and roll a dung ball using its legs. In this paper, we present distributed sensor-driven control for generating dung beetle-like walking and ball rolling behaviors of the developed model. The control mechanism is based on the Walknet bio-inspired controller. Furthermore, this paper gives a proof of concept that locomotion and object manipulation/transportation can be achieved without installing additional manipulators and/or grippers on a robot as shown by other works. To this end, the bio-inspired strategy can avoid the requirement of additional energy to power the manipulator or gripper system; thereby leading to an energy-efficient multi-functional robotic system.

2016

1. Embodied control of a dung beetle inspired hexapod M. Thor, and T. Strøm-Hansen University of Southern Denmark - Campusvej 55, 5230 Odense, Denmark 2016 [PDF]   Bachelor Thesis   Top grade