Robot 2.0: Deep reinforcement learning for industrial robotics

Description

There are only 3 million existing industrial robots in the world. Contrast that with 1 billion cars, and you see why self-driving gets more attention. Yet every other robot manufacturer we talk to says they cannot make robots fast enough. The International Federation of Robotics numbers just showed another year of 30% growth. The commercial opportunity for AI-enabled robots is huge, particularly in the logistics, materials-handling, and automotive industries. These market opportunities are compelling because they are large, have already embraced some measure of automation, and have problem characteristics that are well suited to AI solutions.

At a societal level, AI-enabled robots will address the challenges of outsourcing and the balance of trade, per capita productivity, competitiveness, inequality, and safety—some of the most long-standing and important problems facing the United States today. The best market for this opportunity is ecommerce fulfillment. Estimates of the total size of the materials-handling and logistics market vary, but most agree that it will approach $50 billion globally by 2022 and experience annual growth rates of 20%. After materials handling and logistics, the best use case is the automotive manufacturing industry. Automotive manufacturing is a $95 billion industry within the US, where approximately 10% of the final cost of goods go toward manual labor.

Bastiane Huang examines how AI-enabled robots are used in warehouse automation, including recent progress in deep reinforcement learning, imitation learning, and real world requirements for various industrial problems, and how you can use warehouse robotics as a crystal ball and example for other industries, including manufacturing, food assembly, and surgical robots.