Research Evolutionary Reinforcement Learning Evolutionary reinforcement learning uses the design flexibility of evolutionary algorithms to tackle the most general paradigm of machine learning, reinforcement learning. SUNA is the first and currently the only algorithm that takes this flexibility to the limit, unifying most of the previous proposed features into one unified neuron model.
Motion capture and ground reaction force measurements in freely walking insects Electrophysiology: Intra- and extracellular recordings of nerves, muscles and individual neurons Neurobionics: Our research focus is on the function of active tactile sensing touch and distributed proprioception the sense of posture.
Our main model systems are stick insects. These nocturnal insects rely heavily upon sensory feedback to control locomotion. Stick insects have been used as study organisms in animal physiology and neurobiology for about a century - accordingly, we can build on a lot of detailed knowledge regarding anatomy, physiology and behaviour.
Methodologically, their relatively low speed and large size makes them amenable to many electrophysiological techniques, and allows detailed analyses of natural, unrestrained movement behaviour.
In a parallel research effort we study how stick insects integrate feedback from load sensors on their legs campaniform sensilla to control locomotion. In a complementary approach, we study how individual leg segments are loaded and unloaded in freely walking animals.
Here we take advantage of the relatively long legs of stick insects to simultaneously record 3D joint kinematics, single leg ground reaction forces and multi-channel electromyograms. Campaniform sensilla on the tibia of a stick insect. We use detailed natural whole-body movement data of walking stick insects to model sensory feedback from different proprioceptors on the legs.
How well are body posture and movements estimated from distributed proprioceptors alone? Does an internal body model, which takes the geometry of the animal into account, improve purely sensor-based estimates? Schematic of a hairfield model. As a consequence, many stick insect species use their antennae or feelers for tactile near-range orientation.
During locomotion, they actively move their antennae as if to search the ambient space for obstacles. Once an antenna touches an object, the animal typically responds to this contact event by directed movements. For example, it may move a front leg to reach for the obstacle touched.
|Book Chapters||This holds for natural organisms as well as for robots.|
|Research Faculty | COLLEGE OF ENGINEERING||History[ edit ] The name biomimetics was coined by Otto Schmitt in the s. The term bionics was coined by Jack E.|
|Choose your preferred view mode||Design and Control of continuum octopus-like robotic manipulators. The polychaete annelid marine worms propel themselves in a variety of challenging locomotion environments by a unique form of tail-to-head body undulations, combined with the synchronized action of numerous parapodial lateral appendages.|
In a number of recent and ongoing projects, we study i how the animals adapt their tactile exploration behaviour to spontaneous and induced changes in speed and walking direction, ii whether and how the tactile contact sequence predicts the forthcoming movements, and iii how the coordinate-transfer may work by which the antenna "tells" the leg when and where to grasp an obstacle.
Movement trajectory of a stick insect antenna during walking. To understand what is encoded in the antennal mechanosensory system and how behaviourally relevant information is transferred to the legs, we study descending interneurons that mediate tactile and proprioceptive information from the antennae to thoracic neural networks.
In particular, we focus on the role of an identified giant interneuron, cONv, as the animal interacts with its environment.
Owing to its specific properties cONv can be identified by means of extracellular nerve recordings. In this project we develop a bio-inspired CPG-based control model of the stick insect antenna. Using both simulated and robotic antennae we study how antennal movement strategies are integrated with sensory information such as antennal vibration to localise objects effectively in 3D space.
Model of a bio-inspired antenna.
The tracking system is being designed to investigate the role of antennal movements in active tactile exploration, pattern recognition and tactile learning. Top view of a honeybee with tracked antennae overlaid in green.Mar 02, · Naomi Ehrich Leonard Professor Mechanical and Aerospace Engineering, Princeton University Friday, March 2, Bio-inspired dynamics for multi-agent decision-making.
Whether for good or ill, robotics has been a prime application venue for bio-inspired design. More organic devices are being developed at an increasing rate because of progress in our ability to process information, make things smaller and use new materials with novel properties.
Multi-modal sensing (Radar, Lidar, Imager, etc) lifting, speech, cognition, and basically anything a human can do. Many of today's robots are inspired by nature, contributing to the field of bio-inspired robotics.
Deep learning is the science of training large artificial neural networks. Deep neural networks. Among them, thanks to many advantages in parallel distributed structure, nonlinear mapping, ability to learn from examples, high generalization performance, and capability to approximate an arbitrary function with sufficient number of neurons, the neural-network-based approach is a competitive way to control movements of robot manipulators.
Firstly, based on the establishment of bio-inspired neural network model, AUV working environment is represent by it, there is one-to-one correspondence between each neuron in neural network and the position of the grid map of underwater environment. Bio-inspired Robotics Biomechanics Biomedical Robots Biomimetic Robotics Computer Vision Distributed Sensing Motion Planning and Scheduling Multi-Agent Systems Multimedia Robotics Multi-Robots Neural Computation Neural Networks and .