The Watershed Moment for Humanoid Robotics?

Amazon has a long history of adopting robotics in pursuit of efficiency improvements—it currently deploys over 750,000 robots of various form factors and is actively looking to broaden its portfolio. In October 2023, Amazon announced its intended trial of Agility Robotics’ Digit humanoid robot—an emerging type of Autonomous Mobile Robot (AMR)—in Amazon’s robotics Research and Development (R&D) facility near Seattle, Washington (United States). The robots will initially be tasked with moving empty boxes, but the trial deployment is a test of Agility Robotics’ value proposition and does not guarantee broader adoption. The Digit robot is a 5-foot, 9-inch tall humanoid, a form factor anthropomorphic in appearance and theoretically capable of performing the diversity of manual tasks a human worker can. In September, Agility Robotics announced the construction of RoboFab—the world’s first humanoid robot factory—a facility capable of producing 10,000 humanoids a year. A growing number of companies are pursuing humanoid robotics, including automotive giants Hyundai, Toyota, and Tesla. Other companies specifically targeting warehousing and logistics include 1X Technologies, Apptronik, Figure AI, and UBTech. Although the technology has long been a science-fiction trope—and the economic value of humanoid AMRs is yet to be quantified—both governments and private companies are increasing investments in the belief that humanoid AMRs may be the catch-all to address both aging populations and tightening labor markets.

Agility Robotics’ offering is impressive progress for a technology that has, until now, been confined to research. Stakeholders active in the space are attempting to build the Holy Grail of robotics—a general-purpose robot. This is a tall order that many hope to accomplish with the incorporation of generative Artificial Intelligence (AI) and the adaptability this affords. Expectations for humanoids are the same adaptability as a human worker. Ideally, humanoids will fit seamlessly into any process without the need for knowledge of the greater ecosystem; changing tasks from picking and packaging to loading vehicles with a simple verbal command. Effectively achieving this with the 100% success rate the enterprise sector demands will be a very difficult feat. In its current state, the Digit AMR appears to move much slower and more cautiously than human workers, and it can only perform limited, non-technical operations that appear to be preprogrammed, rather than interpreted through a generative AI controller. In one extreme example at a trade show event, a Digit failed completely and collapsed to the ground before onlookers.

Market sectors exist where the value of humanoid AMRs could be exceptional. For example, a commonly sought application is employing robots as carers for vulnerable people in human-centric environments such as homes and hospitals. Further, a rugged bipedal robot is well suited to operating in brownfield settings such as mines and rural or unstructured environments. This creates significant opportunity for military applications, along with raw material extraction, last-mile delivery, and vertical deployment within both the construction and agricultural sectors. Despite its potential, humanoid AMR technology risks finding itself in the same quagmire as autonomous cars. Humanoids will be expected to be nigh infallible and pose zero risk to the people they operate around. The level of situational awareness—and by extension, compute—that such a solution would require is simply not practically viable in most environments. Removing human unpredictability reduces risk, however. Lights-out warehousing would be an ideal deployment for humanoid AMRs and is feasibly the end state of the Amazon and Agility Robotics partnership. In a further analogy to autonomous vehicles, the industry may broadly adopt a tiered autonomy hierarchy for humanoid AMRs: Level 5 autonomy may be more adept than a skilled human worker, featuring human-like comprehension of verbal commands, and offering total safety when around human workers; Level 0 may be the bipedal equivalent of a dumbwaiter; and Digit, in its current state, would likely be said to possess Level 2 autonomy—able to perform basic tasks with some degree of awareness of its environment. Quantifying performance expectations in this way will help chart the technology’s development and aid adoption.

Agility Robotics’ partnership with Amazon will be the first true test of the commercial viability of the humanoid form factor in its current state of development. Success will signal to the market that the technology is mature enough for wide-scale enterprise adoption but, even if the gambit fails, this type of AMR will continue to draw much attention and further progress is inevitable. Practically, the opportunity cost of investment must be weighed. Replacing low-paid manual workers with an expensive AMR—a previous model of Digit was priced around the US$250,000 mark—may not immediately produce the vast Return on Investment (ROI) that proponents envision. Agility Robotics’ Digit, in its current form, is only capable of performing the most rudimentary of tasks at a slow pace. Would it be cost effective to replace eight manual workers with one AMR? Reduced downtime and 24-hour operation work in the technology’s favor. The promise of the technology also bodes well for Robots-as-a-Service (RaaS) business models that are regularly hampered by the complexity of integrating robotics solutions into existing value chains. Stakeholders would do well to closely watch the development of humanoid robotics, which has the potential to become a booming sector. Amazon may be about to fire the starting gun.