Model-free deep reinforcement learning algorithms have been shown to be capable of learning a wide range of robotic skills, but typically require a very large number of samples to achieve good performance. Model-based algorithms, in principle, can provide for much more efficient learning, but have proven difficult to extend to expressive, high-capacity models such as deep neural networks. In this work, we demonstrate that medium-sized neural network models can in fact be combined with model predictive control (MPC) to achieve excellent sample complexity in a modelbased reinforcement learning algorithm, producing stable and plausible gaits to accomplish various complex locomotion tasks. We also propose using deep neural network dynamics models to initialize a model-free learner, in order to combine the sample efficiency of model-based approaches with the high taskspecific performance of model-free methods. We empirically demonstrate on MuJoCo locomotion tasks that our pure modelbased approach trained on just random action data can follow arbitrary trajectories with excellent sample efficiency, and that our hybrid algorithm can accelerate model-free learning on high-speed benchmark tasks, achieving sample efficiency gains of 3 − 5× on swimmer, cheetah, hopper, and ant agents. Videos can be found at https://sites.google.com/view/mbmf

Mobility and robustness are two important features for practical applications of robots. Soft robots made of polymeric materials have the potential to achieve both attributes simultaneously. Inspired by nature, this research presents soft robots based on a curved unimorph piezoelectric structure whose relative speed of 20 body lengths per second is the fastest measured among published artificial insect-scale robots. The soft robot uses several principles of animal locomotion, can carry loads, climb slopes, and has the sturdiness of cockroaches. After withstanding the weight of an adult footstep, which is about 1 million times heavier than that of the robot, the system survived and continued to move afterward. The relatively fast locomotion and robustness are attributed to the curved unimorph piezoelectric structure with large amplitude vibration, which advances beyond other methods. The design principle, driving mechanism, and operating characteristics can be further optimized and extended for improved performances, as well as used for other flexible devices.

No abstract

Flapping wings provide unmatched maneuverability for flying micro-robots. Recent advances in modelling insect aerodynamics show that adequate wing rotation at the end of the stroke is essential for generating adequate flight forces. W e developed a thorax structure using four bar frames combined with an extensible fan-fold wing to provide adequate wing stroke and rotation. Flow measurements on a scale model of the beating wing show promising aerodynamics. Calculations using a simple resonant mechanical circuit model show that piezoelectric actuators can generate SUBcient power, force and stroke to drive the wings at 150 Hz.

A 2 D 0 F res on an t t horax s t T U ct u re signed and fabricated for the MFI project. Miniature piezoelectric PZN-PT unimorph actuators were fabricated and used to drive a four-bar transmission mechanism. T h e current tho,rax design utilizes two actuated four-bars and a spherical joint t o drive a rigid wing. Rotationally compliant flexure joints have been tested with lifetimes over lo6 cycles. Wing spars were instrumented with strain gauges for force measurement and closed-loop wing control.