Sensorized Skin With Biomimetic Tactility Features Based on Artificial Cross-Talk of Bimodal Resistive Sensory Inputs.

Tactility in biological organisms is a faculty that relies on a variety of specialized receptors. The bimodal sensorized skin, featured in this study, combines soft resistive composites that attribute the skin with mechano- and thermoreceptive capabilities. Mimicking the position of the different natural receptors in different depths of the skin layers, a multi-layer arrangement of the soft resistive composites is achieved. However, the magnitude of the signal response and the localization ability of the stimulus change with lighter presses of the bimodal skin. Hence, a learning-based approach is employed that can help achieve predictions about the stimulus using 4500 probes. Similar to the cognitive functions in the human brain, the cross-talk of sensory information between the two types of sensory information allows the learning architecture to make more accurate predictions of localization, depth, and temperature of the stimulus contiguously. Localization accuracies of 1.8 mm, depth errors of 0.22 mm, and temperature errors of 8.2 °C using 8 mechanoreceptive and 8 thermoreceptive sensing elements are achieved for the smaller inter-element distances. Combining the bimodal sensing multilayer skins with the neural network learning approach brings the artificial tactile interface one step closer to imitating the sensory capabilities of biological skin.

Autonomous detection and sorting of litter using deep learning and soft robotic grippers.

Road infrastructure is one of the most vital assets of any country. Keeping the road infrastructure clean and unpolluted is important for ensuring road safety and reducing environmental risk. However, roadside litter picking is an extremely laborious, expensive, monotonous and hazardous task. Automating the process would save taxpayers money and reduce the risk for road users and the maintenance crew. This work presents LitterBot, an autonomous robotic system capable of detecting, localizing and classifying common roadside litter. We use a learning-based object detection and segmentation algorithm trained on the TACO dataset for identifying and classifying garbage. We develop a robust modular manipulation framework by using soft robotic grippers and a real-time visual-servoing strategy. This enables the manipulator to pick up objects of variable sizes and shapes even in dynamic environments. The robot achieves greater than 80% classified picking and binning success rates for all experiments; which was validated on a wide variety of test litter objects in static single and cluttered configurations and with dynamically moving test objects. Our results showcase how a deep model trained on an online dataset can be deployed in real-world applications with high accuracy by the appropriate design of a control framework around it.

Emergence of behavior through morphology: a case study on an octopus inspired manipulator.

The complex motion abilities of the Octopus vulgaris have been an intriguing research topic for biologists and roboticists alike. Various studies have been conducted on the underlying control architectures employed by these high dimensional biological organisms. Researchers have attempted to replicate these architectures on robotic systems. Contrary to previous approaches, this study focuses on a robotic system, which is only morphologically similar to the Octopus vulgaris, and how it would behave under different control policies. This sheds light on the underlying optimality principles that these biological systems employ. Open loop control policies are obtained through a trajectory optimization method on a learned forward dynamic model. The motion patterns emerging from variations in morphology and environment were then derived to study the role of the body and environment. Results show that for the specific case of dynamic reaching with a soft appendage, the invariance in motion profile is a fundamental constraint imposed by the morphology and environment, independent from the controller. This suggests how morphological design can simplify stable control even for highly dimensional nonlinear dynamical systems and can provide insights into design of new soft robotic mechanisms.

Soft robot perception using embedded soft sensors and recurrent neural networks.

Recent work has begun to explore the design of biologically inspired soft robots composed of soft, stretchable materials for applications including the handling of delicate materials and safe interaction with humans. However, the solid-state sensors traditionally used in robotics are unable to capture the high-dimensional deformations of soft systems. Embedded soft resistive sensors have the potential to address this challenge. However, both the soft sensors-and the encasing dynamical system-often exhibit nonlinear time-variant behavior, which makes them difficult to model. In addition, the problems of sensor design, placement, and fabrication require a great deal of human input and previous knowledge. Drawing inspiration from the human perceptive system, we created a synthetic analog. Our synthetic system builds models using a redundant and unstructured sensor topology embedded in a soft actuator, a vision-based motion capture system for ground truth, and a general machine learning approach. This allows us to model an unknown soft actuated system. We demonstrate that the proposed approach is able to model the kinematics of a soft continuum actuator in real time while being robust to sensor nonlinearities and drift. In addition, we show how the same system can estimate the applied forces while interacting with external objects. The role of action in perception is also presented. This approach enables the development of force and deformation models for soft robotic systems, which can be useful for a variety of applications, including human-robot interaction, soft orthotics, and wearable robotics.

Learning dynamic models for open loop predictive control of soft robotic manipulators.

The soft capabilities of biological appendages like the arms of Octopus vulgaris and elephants' trunks have inspired roboticists to develop their robotic equivalents. Although there have been considerable efforts to replicate their morphology and behavior patterns, we are still lagging behind in replicating the dexterity and efficiency of these biological systems. This is mostly due to the lack of development and application of dynamic controllers on these robots which could exploit the morphological properties that a soft-bodied manipulator possesses. The complexity of these high-dimensional nonlinear systems has deterred the application of traditional model-based approaches. This paper provides a machine learning-based approach for the development of dynamic models for a soft robotic manipulator and a trajectory optimization method for predictive control of the manipulator in task space. To the best of our knowledge this is the first demonstration of a learned dynamic model and a derived task space controller for a soft robotic manipulator. The validation of the controller is carried out on an octopus-inspired soft manipulator simulation derived from a piecewise constant strain approximation and then experimentally on a pneumatically actuated soft manipulator. The results indicate that such an approach is promising for developing fast and accurate dynamic models for soft robotic manipulators while being applicable on a wide range of soft manipulators.