General Purpose Low-Level Reinforcement Learning Control for Multi-Axis Rotor Aerial Vehicles.

This paper proposes a multipurpose reinforcement learning based low-level multirotor unmanned aerial vehicles control structure constructed using neural networks with model-free training. Other low-level reinforcement learning controllers developed in studies have only been applicable to a model-specific and physical-parameter-specific multirotor, and time-consuming training is required when switching to a different vehicle. We use a 6-degree-of-freedom dynamic model combining acceleration-based control from the policy neural network to overcome these problems. The UAV automatically learns the maneuver by an end-to-end neural network from fusion states to acceleration command. The state estimation is performed using the data from on-board sensors and motion capture. The motion capture system provides spatial position information and a multisensory fusion framework fuses the measurement from the onboard inertia measurement units for compensating the time delay and low update frequency of the capture system. Without requiring expert demonstration, the trained control policy implemented using an improved algorithm can be applied to various multirotors with the output directly mapped to actuators. The algorithm's ability to control multirotors in the hovering and the tracking task is evaluated. Through simulation and actual experiments, we demonstrate the flight control with a quadrotor and hexrotor by using the trained policy. With the same policy, we verify that we can stabilize the quadrotor and hexrotor in the air under random initial states.

Somatosensory actuator based on stretchable conductive photothermally responsive hydrogel.

Mimicking biological neuromuscular systems' sensory motion requires the unification of sensing and actuation in a singular artificial muscle material, which must not only actuate but also sense their own motions. These functionalities would be of great value for soft robotics that seek to achieve multifunctionality and local sensing capabilities approaching natural organisms. Here, we report a soft somatosensitive actuating material using an electrically conductive and photothermally responsive hydrogel, which combines the functions of piezoresistive strain/pressure sensing and photo/thermal actuation into a single material. Synthesized through an unconventional ice-templated ultraviolet-cryo-polymerization technique, the homogenous tough conductive hydrogel exhibited a densified conducting network and highly porous microstructure, achieving a unique combination of ultrahigh conductivity (36.8 milisiemens per centimeter, 103-fold enhancement) and mechanical robustness, featuring high stretchability (170%), large volume shrinkage (49%), and 30-fold faster response than conventional hydrogels. With the unique compositional homogeneity of the monolithic material, our hydrogels overcame a limitation of conventional physically integrated sensory actuator systems with interface constraints and predefined functions. The two-in-one functional hydrogel demonstrated both exteroception to perceive the environment and proprioception to kinesthetically sense its deformations in real time, while actuating with near-infinite degrees of freedom. We have demonstrated a variety of light-driven locomotion including contraction, bending, shape recognition, object grasping, and transporting with simultaneous self-monitoring. When connected to a control circuit, the muscle-like material achieved closed-loop feedback controlled, reversible step motion. This material design can also be applied to liquid crystal elastomers.