Nuo Ma – Reflection 4


WeBuild: Automatically Distributing Assembly Tasks Among Collocated Workers to Improve Coordination


In the first paper, the goal is to improve group efficiency and coordination. Specifically, participants was asked to complete manual assembly tasks, algorithm will assign subtasks to different collocated workers. They observed increased efficiency through reduction of startup time and overall time for completion. The results shows faster but not significant completion time, but members get started faster when using algorithm distribution  (24 sec vs 240 sec). However, in interview, participants rated themselves less aware of the whole problem.

In the second paper: crowdsourced fabrication, the author tried to achieve the goal of coordinating large number of workers to construct a pre-designed structure. Specifically, 108 conference participants and staff members was asked to build 12’ tall pavilion, with real time guidance provided through a smartwatch. To measure the results, the author observed the construction process, with user survey, a 56% response rate and onsite interviews. Specifically metrics includes build progress, module completion time, overall rating, perception of whether it was challenging or straight forward, if the device was useful and self-efficiency/ confidence. The result shows that volunteers took average 20 min vs staff take 16 min. Familiarity with the task affects efficiency. However sometimes volunteers did not tighten enough to ensure structural integrity, and sometimes replacement is needed of incorrectly built module.


From my own experience of assembly IKEA furnitures, although i didn’t have algorithms to assist with the process, I find there are different strategies. Sometimes, for larger projects, when I work with my friends, people tend to first start with figuring out the entire process, then choose more repetitive tasks, like assembling components with same steps, then work together on the larger piece. In this piece, hierarchy was automatically formed amongst workers because you need to be aware of what you need from other workers. This explained why in user studies, people spend a long time starting up without algorithms. Although this user does not know very specifically the details of other person’s work, but have a general knowledge of what’s going on. The study did not show significant improvement on overall time, and how this method will work on larger groups and more complex tasks? In my opinion, there is a tipping point that this method works for less complicated project and smaller groups, no significant efficiency different will be shown until the group becomes too large and work becomes too complicated, then workers does not really need to know the whole picture of the project, and focus on more repetitive tasks they were assigned to. But someone at higher level should know how to put these pieces together into a bigger piece, and this is the hierarchy people automatically find mentioned earlier in this discussion. So if there are future studies, instead of seeing users get to choose between repetitive work and more diverse work, I’d like to see how this ‘hierarchical’ task distribution works for larger groups and more complex tasks, and if possible, study if such tipping point exists and how to find them via practical user study.