Packaging Deep RL agents for cross-platform deployment

Although the grandest success of Deep RL has been in the domain of game playing (Atari, Chess, Go, Shogi) and simulated robotics, real-world applications are starting to emerge where Deep RL agents show a lot of promise and value. Deploying Deep RL agents to a variety of physical form factors such as embedded controllers, computers, autonomous cars, drones, and other robots, and so on is expected soon. Differences in hardware processors (CPU, GPU, TPU, FPGA, ASIC), operating systems (Linux, Windows, OSX, Android), architectures (x86, ARM), and form factors (server, desktop, mobile, IoT, embedded systems, and so on) make the deployment process challenging. This recipe includes guidelines around how you can leverage the TensorFlow 2.x framework's ecosystem of libraries, tools, and utilities to package Deep RL agent models suitable for deployments to the web, mobile, IoT, embedded systems, robots, and desktop platforms.

This...