In the race to meet its highly ambitious 5,000 cars per week target in 2017, Tesla relied on extreme automation with more than 1,000 robots for its Model 3 production. However, the production didn't go according to plan, and the company had to reset its target to 2,500 within a year of the initial announcement. In the middle of the "production hell" period, as cited by the company's famous CEO Elon Musk, the company was far behind its schedule. It produced only 2,000 cars per week on average in April 2018, which led Elon to tweet that the company had relied on too many robots and that humans were underrated. The company then quickly changed its strategy and hired workers to do things that robots couldn't and finally met its target in July 2018. While we first wrote about this case study in 2020, here, we analyze what went wrong, now that we have additional details from Tesla.
USE CASE AND BUSINESS IMPACT
One of Tesla's robots was having issues with placing noise-dampening fiberglass fluff on top of a battery pack because of its inability to pick up pieces of fluff. Similarly, another robot that was designed to fasten seats to the car body was struggling to work with bolts and screws that weren't perfectly aligned. Because the overall production efficiency in manufacturing is a multiplication of the efficiencies of individual tasks, even a 1% reduction in efficiency caused by such struggling automation systems could result in a huge cumulative delay. To remedy the situation, the company hired hundreds of human workers every week for a couple of months to perform these tasks, in addition to other tasks like painting and sanding that were causing quality-related issues.
Based on the details later shared by Tesla, the company's speed bump with automation can be summarized as an event caused by relying on low-maturity technology and assuming that manufacturing conditions are always ideal. While there are now technologies like generative adversarial networks (GANs) and synthetic data that can generate thousands of images of improperly aligned nuts and bolts to train computer vision algorithms, robotic grippers are still not able to replicate human dexterity in handling small/light materials. Another point to consider in Tesla's reverse automation case study is that Tesla is often framed as a tech company that mass-produces cars. Traditional automotive manufacturers like GM, Toyota, and Honda are typically very cautious about automating a manual task without running hundreds of simulations and tests to identify any errors. Tesla's imperfect automation strategy emphasizes that before attempting automation, it is critical to understand the timeline for automation of individual tasks, which depends on several factors, such as technology maturity and drivers and barriers for automation. Clients should use the framework in our recently published report on lights-out manufacturing to see when they can expect manufacturing tasks to be fully automated.