Curriculum learning with Hindsight Experience Replay for sequential object manipulation tasks.

Learning complex tasks from scratch is challenging and often impossible for humans as well as for artificial agents. Instead, a curriculum can be used, which decomposes a complex task - the target task - into a sequence of source tasks. Each source task is a simplified version of the next source task with increasing complexity. Learning then occurs gradually by training on each source task while using knowledge from the curriculum's prior source tasks. In this study, we present a new algorithm that combines curriculum learning with Hindsight Experience Replay (HER), to learn sequential object manipulation tasks for multiple goals and sparse feedback. The algorithm exploits the recurrent structure inherent in many object manipulation tasks and implements the entire learning process in the original simulation without adjusting it to each source task. We test our algorithm on three challenging throwing tasks in simulation and show significant improvements compared to vanilla-HER.

Robust Flexible Pressure Sensors Made from Conductive Micropyramids for Manipulation Tasks.

Flexible pressure sensors that can robustly mimic the function of slow-adapting type I (SA-I) mechanoreceptors are essential for realizing human-like object manipulation in artificial intelligent (AI) robots or amputees. Here, we report a straightforward approach to highly sensitive and robust flexible pressure sensors with fast response time and low operating voltage based on conductive micropyramids made of polydimethylsiloxane/carbon nanotube composites. Both numerical simulations and experimental studies show that the pressure-sensing properties of the devices can be systematically tuned by the spatial arrangement of micropyramids. In particular, by tailoring the ratio between the spacing and the pyramidal base length, the optimal pressure sensors can be achieved with a combination of high sensitivity in both low-pressure (<10 kPa) and medium-pressure (10-100 kPa) regimes, rapid response, high mechanical robustness, low operating voltage, and low power consumption, along with linear response and low hysteresis in the medium-pressure regimes. The optimized pressure sensor is further used for constructing a wearable pressure-sensing system that can convert the amplitude of pressure to wirelessly transmittable frequency signals (spikes) with nearly linear response, closely mimicking SA-I mechanoreceptors. Furthermore, we demonstrate that the high uniformity and scalability of the pressure sensors enable large-area pressure-sensing arrays for spatially resolved pressure mapping.

Tactile Sensing and Control of Robotic Manipulator Integrating Fiber Bragg Grating Strain-Sensor.

Tactile sensing is an instrumental modality of robotic manipulation, as it provides information that is not accessible via remote sensors such as cameras or lidars. Touch is particularly crucial in unstructured environments, where the robot's internal representation of manipulated objects is uncertain. In this study we present the sensorization of an existing artificial hand, with the aim to achieve fine control of robotic limbs and perception of object's physical properties. Tactile feedback is conveyed by means of a soft sensor integrated at the fingertip of a robotic hand. The sensor consists of an optical fiber, housing Fiber Bragg Gratings (FBGs) transducers, embedded into a soft polymeric material integrated on a rigid hand. Through several tasks involving grasps of different objects in various conditions, the ability of the system to acquire information is assessed. Results show that a classifier based on the sensor outputs of the robotic hand is capable of accurately detecting both size and rigidity of the operated objects (99.36 and 100% accuracy, respectively). Furthermore, the outputs provide evidence of the ability to grab fragile objects without breakage or slippage e and to perform dynamic manipulative tasks, that involve the adaptation of fingers position based on the grasped objects' condition.