- Robotic Locomotion and Control
- Reinforcement Learning in Robotics
- Prosthetics and Rehabilitation Robotics
- Modular Robots and Swarm Intelligence
- Robot Manipulation and Learning
- Muscle activation and electromyography studies
- Biomimetic flight and propulsion mechanisms
- Evolutionary Algorithms and Applications
- Soft Robotics and Applications
- Robotics and Sensor-Based Localization
- Adversarial Robustness in Machine Learning
- Advanced Sensor and Energy Harvesting Materials
- Gaussian Processes and Bayesian Inference
- Advanced Multi-Objective Optimization Algorithms
- Optimization and Search Problems
- Motor Control and Adaptation
- Robotic Path Planning Algorithms
- Advanced Neural Network Applications
- Robotic Mechanisms and Dynamics
- Evolutionary Game Theory and Cooperation
- Supply Chain and Inventory Management
- Machine Learning and Data Classification
- Robotics and Automated Systems
- Machine Learning and Algorithms
- Bat Biology and Ecology Studies
ShanghaiTech University
2017-2022
Worcester Polytechnic Institute
2022
Google (United States)
2018
University of Cambridge
2016-2017
University of KwaZulu-Natal
2017
University of Cyprus
2016
Osaka University
2012-2015
Hokkaido University
2010
Evolutionary algorithms have previously been applied to the design of morphology and control robots. The space for such tasks can be very complex, which prevent evolution from efficiently discovering fit solutions. In this article we introduce an evolutionary-developmental (evo-devo) experiment with real-world It allows robots grow their leg size simulate ontogenetic morphological changes, is first time that has performed in physical world. To test diverse robot morphologies, legs variable...
This paper presents the design, analysis, and performance evaluation of an omnidirectional transformable wheel-leg robot called OmniWheg. We design a novel mechanism consisting separable omni-wheel 4-bar linkages, allowing to transform between omni-wheeled legged modes smoothly. In wheeled mode, can move in all directions efficiently adjust relative position its wheels, while it overcome common obstacles such as stairs steps. Unlike other articles studying whegs, this implementation with...
While quadruped robots usually have good stability and load capacity, bipedal offer a higher level of flexibility / adaptability to different tasks environments. A multi-modal legged robot can take the best both worlds. In this paper, we propose locomotion framework that is composed hand-crafted transition motion learning-based controller—learnt by novel algorithm called Automated Residual Reinforcement Learning. This aims endow arbitrary with ability walk bipedally. particular, 1) design an...
Conventionally, robot morphologies are developed through simulations and calculations, different control methods applied afterwards. Assuming that predictions simplified representations of our reality, how sure can roboticists be the chosen morphology is most adequate for possible choices in real-world? Here we study influence design parameters creation a with Bayesian morphology-control (MC) co-optimization process. A autonomously creates child robots from set uses Optimization (BO) to...
In this paper, we describe the development of quadruped robot "Ken" with minimalistic and lightweight body design for achieving fast locomotion. We use McKibben pneumatic artificial muscles as actuators, providing high frequency wide stride motion limbs, also avoiding problems overheating. conducted a preliminary experiment, finding out that can swing its limb over 7.5 Hz without amplitude reduction, nor heat problems. Moreover, realized several steps bouncing gait by using simple CPG-based...
Feline locomotion combines great acrobatic proficiency, unparalleled balance and higher accelerations than other animals. Capable of accelerating from 0 to 100 km h <sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">-1</sup> in three seconds, the cheetah (Acinonyx jubatus) is still a mystery which intrigues scientists. Aiming for better understanding source such speeds, we develop biomimetic platform, where musculoskeletal parameters (range motion...
Soft grippers significantly widen the palpation capabilities of robots, ranging from soft to hard materials without assistance cameras. From a medical perspective, detection size and shape inclusions concealed within three-dimensional (3D) objects is meaningful for early cancer through palpation. This article proposes framework variable-stiffness object recognition using tactile information collected by force sensitive resistors on three-finger gripper. A 15 × 50 spatiotemporal image...
To maintain balance during dynamic locomotion, the effects of proprioceptive sensory feedback control (e.g. reflexive control) should not be ignored because its simple sensation and fast reaction time. Scientists have identified pathways reflexes; however, it is difficult to investigate their locomotion controlled by a complex neural system current technology does allow us change in living humans. understand these effects, we construct musculoskeletal bipedal robot, which has similar body...
Open-loop control strategies for legged robot locomotion have been investigated by many researchers because of the advantages in terms simplicity and robustness, although influence inputs to performance is not fully clarified. This paper investigates two most basic forms input, sinusoidal pulsed signals, be used a class hopping based on parallel elastic actuation. Our results show that torque outperforms its counterpart with lower energy expenditure. Moreover, driving capable keeping same...
Although three million people around the world suffer from lack of one or both upper limbs 80% this number is located within developing countries. While prosthetic prices soar with technology 3D printing and low cost electronics present a sensible solution for those that cannot afford expensive prosthetics. The electronic control design low-cost hand, Touch Hand II, discussed. This paper shows sensorless techniques can be used to reduce complexities, costs, provide easier access electronics....
When humans hop, attitude recovery can be observed in both the sagittal and frontal planes. While it is agreed that brain plays an important role leg placement, of low-level feedback (the stretch reflex) on plane stabilization remains unclear. Seeking to better understand contribution soleus reflex rolling stability, we performed experiments a biomimetic humanoid hopping robot. Various responses touching floor, ranging from no response long muscle activations, were examined, effect delay...
Studies on decerebrate walking cats have shown that phase transition is strongly related to muscular sensory signals at limbs. To further investigate the role of such terminating stance phase, we developed a biomimetic feline platform. Adopting link lengths and moment arms from an Acinonyx jubatus, built pair hindlimbs connected hindquarter attached it sliding strut, simulating solid forelimbs. Artificial pneumatic muscles simulate biological through control method based EMG (Felis catus)....
Adaptability of quadruped animals is not solely reached by brain control, but the interaction between its body, environment, and control. Especially, morphology body supposed to contribute largely adaptability. We have tried understand quadrupedal locomotion building a bio-inspired robot named "Pneupard", which has feline-like muscular-skeletal structure. In our previous study, we successfully realized alternative gait hindlimbs reflex control based on sole touch information, called an...
Malfunctions on industrial robots can cost factories 22000 dollars per minute. Although the benefits of a fault-tolerant robot arm are clear, redundant sensors would steeply add to costs such while machine learning-based methods spend too much time learning robot's model. We propose simple but highly effective method infer which joint underwent failure and at angle this is constrained to, then, modify inverse kinematics (IK) algorithm adaptively achieve goal. Our involves combining with QR...
Cats, from tiny domestic cats (Felix Catus) to big tigers (Panthera Tigris), are well known for their great acrobatic skills and hunting ability. Aiming better understand how the feline family interacts with environment, we adopt a biomimetic approach on hopping hindlimb. Using air muscles simulate compliance of biological muscles, this robotic hindlimb has seven changes direction. We individually evaluate estimate contribution jumping Finally, successfully control direction using non-linear...
The Spring-Loaded Inverted Pendulum (SLIP) model has been in play for many years as the most plausible explanation both walking and running gaits. Although current knowledge on this advanced significantly over years, a more in-depth analysis of vast parameter space was always hindered by inefficient searching tools. Beyond finding stable representation different velocities, an ideal search method also needs to consider gait alternatives quickly optimize its velocity maximum possible. In work...
Proximal policy optimization (PPO) has yielded state-of-the-art results in search, a subfield of reinforcement learning, with one its key points being the use surrogate objective function to restrict step size at each update. Although such restriction is helpful, algorithm still suffers from performance instability and inefficiency sudden flattening curve. To address this issue we present novel functional clipping algorithm, named Policy Optimization Smoothed Algorithm (PPOS), critical...
Autonomous Valet Parking (AVP) in an under- ground garage is emerging smart vehicle solution that the community believes to be solvable with close-to-market sensors. Absence of GPS signals and a high degree self-similarity however render global visual localization such environments highly challenging problem. We present novel underground parking method relies on text recognition wild as well optical character (OCR) automatically detect slot numbers. The detected numbers are then correlated...
With continuous advancements on active prosthetics the detection of user's intention becomes new technological bottleneck. While electromyography (EMG) is widely used to detect individual muscular contributions, sweat and relative sensor movements degrade quality signal over time. In this paper, we bypass problems created with skin contact analyzing activation Archimedean Spiral (AS) electrodes. We compare traditional EMG electrodes AS electrodes, stacked up in textile embroidered layers...
Energy efficiency is critical for the locomotion of quadruped robots. However, energy values found in simulations do not transfer adequately to real world. To address this issue, we present a novel method, named Policy Search Transfer Optimization (PSTO), which combines deep reinforcement learning and optimization create energy-efficient robots The policy search process are performed by TD3 algorithm transferred open-loop control trajectory further optimized numerical methods, conducted on...
Animals have several peripheral feedback control networks such as stretch reflexes that are supposed to be contributing stability of their behavior. In this paper, we focus on rolling hopping a biped robot, and investigate schemes an experimental robot for implementing the reflex: no reflex, undelayed delayed how they contribute stability. For eliminating effect changing environment initial postures, conduct large number experiments real robot. From results, can conclude (1) two reflex...
The articles in this special section focus on the development and manufacture of soft robotics.