Roboticsnavigation learningStory 04
A walker teaches a wheelchair where to go
What changedNortheastern researchers collected 37.2 kilometres of routes using a rollator walker and smartphone. Because the walker cannot climb stairs or uncut curbs, its paths reflect wheel-friendly choices. Models trained with that data transferred to a powered wheelchair in selected scenarios. The collection tools and full dataset are still awaiting release.
The useful part
Why it matters
A cheap physical proxy can collect routes that respect wheel constraints without needing the target robot for every demonstration.
Keep in mind
Good to know
Transfer demonstrations cover curbs, stairs and curb cuts. Code, trained weights and the privacy-processed dataset are not yet public.
Evidence