Policy Compression for Intelligent Continuous Control on Low-Power Edge Devices.

Interest in deploying deep reinforcement learning (DRL) models on low-power edge devices, such as Autonomous Mobile Robots (AMRs) and Internet of Things (IoT) devices, has seen a significant rise due to the potential of performing real-time inference by eliminating the latency and reliability issues incurred from wireless communication and the privacy benefits of processing data locally. Deploying such energy-intensive models on power-constrained devices is not always feasible, however, which has led to the development of model compression techniques that can reduce the size and computational complexity of DRL policies. Policy distillation, the most popular of these methods, can be used to first lower the number of network parameters by transferring the behavior of a large teacher network to a smaller student model before deploying these students at the edge. This works well with deterministic policies that operate using discrete actions. However, many real-world tasks that are power constrained, such as in the field of robotics, are formulated using continuous action spaces, which are not supported. In this work, we improve the policy distillation method to support the compression of DRL models designed to solve these continuous control tasks, with an emphasis on maintaining the stochastic nature of continuous DRL algorithms. Experiments show that our methods can be used effectively to compress such policies up to 750% while maintaining or even exceeding their teacher's performance by up to 41% in solving two popular continuous control tasks.

Synthesis of 1,6/7-(NO2 )2 -Perylenediimide and 1,6/7-(NH2 )2 -Perylenediimide as Regioisomerically Pure Materials.

The use of perylenediimide (PDI) building blocks in materials for organic electronic is of considerable interest. This popular n-type organic semiconductor is tuned by introducing peripheral groups in their ortho and bay positions. Such modifications radically alter their optoelectronic properties. In this article, we describe an efficient method to afford regioisomerically pure 1,6/7-(NO2 )2 - and (NH2 )2 -PDIs employing two key steps: the selective crystallization of 1,6-(NO2 )2 -perylene-3,4,9,10-tetracarboxy tetrabutylester and the nitration of regiopure 1,7-Br2 -PDI with silver nitrite. The optoelectronic properties of the resulting regioisomerically pure dinitro, diamino-PDIs and bisazacoronenediimides (BACDs) are reported and demonstrate the need to separate both regioisomers of such n-type organic semiconductors for their inclusion in advanced optoelectronic devices. For the first time, the two regioisomers of the same PDI starting material are available on the multigram scale, which will stimulate the exploration of regioisomerism/properties relationship for this family of dyes.