Identifying\u00a0an object and picking it up are tasks that still elude modern robotic systems. In 2015 Amazon Robotics started its “Picking and Packing Challenge” to increase focus from researchers and startups on this exact problem. Inviting submissions from teams around the world, Amazon hosts the annual challenge and offers cash prizes to teams whose robots can quickly and accurately pick specific items (components of a hypothetical order) from boxes in one area, and place these items in a box-to-be-shipped (for a hypothetical customer). As the cost of labor, the human pick & packers, increases and Amazon ships more products to more places, the benefits of automating the picking and packing process are increasingly lucrative for the company. While its proprietary logistics and order fulfillment technology enabled Amazon’s rise over the last twenty years, Amazon does not presume to have the best way to solve the same problems in the future. The “Picking and Packing Challenge” gives Amazon Robotics access to completely new approaches, bright minds, and real solutions.<\/p>\n
The task of identifying and picking up objects may be simple for a small human to complete (the human hand has 27 degrees of freedom), but for a robot actually prove quite challenging. The task is particularly complex because in the case of Amazon order fulfillment, the object of interest can be any size, weight, shape or material. For a robot to excel at the task, it must have the ability to perceive an object in its environment, plan its own movement, attempt physical manipulation, and detect and correct errors in packing the final order. Thus, this problem requires expertise in a number of fields: visual detection (software), robotic control algorithms (software) and gripping techniques (hardware). The choice to crowdsource ideas, knowledge, and implementation from teams around the world is essential for Amazon to stay competitive. Universities, where research teams work in many diverse fields, as well as startups who solve similar challenges, are much more likely to offer compelling solutions than an in-house innovation team, specifically because this task sits at the convergence of multiple fields. These are fields that are still developing, as well, thus Amazon is unlikely to find the best approach among employees it has already hired and removed from the world of research.<\/p>\n
Amazon\u2019s short-term solution for solving the future of fulfillment and logistics is to host the open challenge. In the long-term, Amazon either hires talented engineers and scientists who compete in the challenge or acquires entire teams (or even businesses). Hiring and acquisition guarantee that Amazon finds and retains the best talent and ideas solving these exact problems today.<\/p>\n
It is unclear, however, if the best future solution requires broadening the problem of \u201cPicking and Packing.\u201d A successful \u201cPicking and Packing\u201d robot would directly replace human pickers & packers in Amazon\u2019s warehouse. In the current setup, a worker stands in one place all day, with boxes passing by on a conveyor belt with items needed to fulfill a specific order, and the human only needs to pick up the appropriate amount of items and place them in a box. The assumption is that the logistics and system external to this singular task remain the same: boxes of similar items arranged in the warehouse, coming to one location. However, the best new approaches to warehouse automation may assume an entirely new construct. For example, Ocado (1), a British supermarket, has solved warehouse automation by constructing a grid over which robots can move seamlessly, centralizing the machine or system intelligence outside of the moving robots, and optimizing vertical space.<\/p>\n
<\/p>\n
Two questions remain. First,\u00a0is the assumption of Amazon’s current warehouse setup that is inherent in the way the challenge is structured a good one? Stated another way, would Amazon discover more effective warehouse automation solutions by starting with a blank slate? And secondly, is there an\u00a0asymptote on the graph of “Time to Fulfill an Order” versus “Customer Willingness to Pay \u201d and is Amazon approaching it?<\/p>\n
My assumption in answering the second question is that Amazon would rather solve the problem & find out otherwise, then let competitors in the space advance beyond them.<\/p>\n
<\/p>\n
<\/p>\n
Amazon's "Open Innovation" approach to solving warehouse automation<\/p>\n","protected":false},"author":11547,"featured_media":35449,"comment_status":"open","ping_status":"closed","template":"","categories":[52,21,346,2461,5027,5028,2663],"class_list":["post-35448","hck-submission","type-hck-submission","status-publish","has-post-thumbnail","hentry","category-amazon","category-logistics","category-machine-learning","category-machine-vision","category-packing","category-picking","category-robots","hck-taxonomy-organization-amazon","hck-taxonomy-industry-shipping","hck-taxonomy-country-united-states"],"connected_submission_link":"https:\/\/d3.harvard.edu\/platform-rctom\/assignment\/rc-tom-challenge-2018\/","yoast_head":"\n