Starship: A case study for accessibility in robotics

I initially wrote the following in 2021 as part of a DEI statement for faculty applications. However, now that I have my own website, I think this content is more appropriate here as a short blog post. I feel especialy close to this story because the events are located in Pittsburgh and because I have been acquainted with Emily for some time. I have known Emily since we were both students RPI. We each also graduated in the same class before moving to Pittsburgh to start our PhDs.

Sidewalk hazards

Researchers in computer science and robotics can better recognize relevance of accessibility to research and practice. In 2019, Starship Technologies began operating a fleet of delivery robots around the University of Pittsburgh campus. These robots are smaller than a cubic meter and operate on walkways like pedestrians. In October of that year, a Starship delivery robot stopped and blocked an access ramp at an intersection near the campus. While I might have walked past the robot and forgotten the obstruction, Emily Ackerman, who has a mobility impairment and uses an electric wheelchair, was trapped by the robot blocking her path.¹ Emily was a University of Pittsburgh graduate student and now has completed a Ph.D. in Chemistry; she was a classmate and acquaintance from my alma mater; and she is disability activist. Emily shared this story on social media and in the linked article, chastening Starship for not considering people such as her when they designed their systems and exhorting them to do better. This started a vital conversation with Starship and my community. Yet, the initial response by Starship was dismissive: “Starship reviewed the footage and confirmed that Emily was able to access passage onto the sidewalk.” ² I think Starship’s failure to account for the individuals their robots would share sidewalks with is notable, and this is evident in how Starship communicates and presents itself.

Comparison: Starship and Nuro

I would like go into more depth on this example by comparing two autonomous delivery companies, Starship and Nuro, in terms of how they evaluate and communicate the impacts of their systems. Even after a few years, Starship does not communicate clearly or in detail on accessibility and the impacts of their robots: an “FAQ” on their website discusses topics like safety primarily in terms of their robots’ ability to avoid collisions and harm due to small size and low speeds. However, they omit obvious topics such as mitigating crowding on walkways, plans for accessibility to wheelchair users, or discussion of how they addressed Emily’s safety concerns. More recently, a video of several of Starship’s robots blocking a snowy walkway in Estonia illustrates the company’s apparent lack of consideration or planning for the conditions their robots encounter. Nuro is a somewhat larger company developing autonomous delivery robots for operation in roadways. While the onus to discuss safety is more pressing, their website provides much more extensive discussion of how their robots will impact local communities, the role of safety in planning and design, and accessibility to wheelchair users. What I see here is that significant differences in values are evident from even casual engagement with either company. I believe researchers can emulate companies like Nuro by evaluating the impacts of their research with depth and nuance and by communicating on these topics clearly and openly.