Nick Walker

@[email protected]

[Nick Walker @nickwalker_us]

We got a robot that actually does something

[Nick Walker @nickwalker_us]

Work with Amal Nanavati, @lee_taber, @mavrojean, @leilatakayama, @mayacakmak, @siddhss5.

And Kuri.

[Nick Walker @nickwalker_us]

And if you know someone who's eating dinner right now, maybe send them the video we made for #hri2022: https://t.co/YkzVUrm23I

[Nick Walker @nickwalker_us]

If you want to play with a wandering behavior, I put a little live demo in this post:
https://t.co/Pl1jroezk5

[Nick Walker @nickwalker_us]

Deploying mobile robots is hard, so we don't do it that often. But how else are we going to start to understand how people interact with them? 🧵#hri2022

[Nick Walker @nickwalker_us]

Otherwise, we give some guidance on making systems for other robots in the paper:
https://t.co/SLELqTdfoJ

[Nick Walker @nickwalker_us]

So, get your robots out there! If your lab has a Kuri, you can try our system out-of-the-box:
https://t.co/LtVNshvPH5

[Nick Walker @nickwalker_us]

Obviously wandering alone doesn't cut it for some kinds of tasks. But there might be ways you could make it work in combination with a bit of human help. For low-cost robots, wandering might be the only way you can make a long deployment feasible.

[Nick Walker @nickwalker_us]

The floor the robot ran on was filled with glass walls, shiny metal (cough), and black chain-link banister that barely showed up on lidar until it was a foot away. No fatal cliffs, fortunately.

[Nick Walker @nickwalker_us]

Each of the 12 times it needed to charge, it pinged a helper via a chatbot and they placed it back on the charger.

[Nick Walker @nickwalker_us]

We deployed the wandering system in our building for 32 hours across 4 days. Most of the time its recovery behaviors kept it from getting stuck, but it did need manual rescue 4 times.

[Nick Walker @nickwalker_us]

We found that moving in a direction until obstructed, then selecting a new obstacle-free direction **close to the previously selected direction** was surprisingly effective at getting Kuri to circulate our halls.

[Nick Walker @nickwalker_us]

Wandering's an old idea, and you might have even owned a wandering robot at some point (thank you @rodneyabrooks). Go in a direction until something makes you select a new one, repeat.

[Nick Walker @nickwalker_us]

We realized that, for what we wanted to study, it was okay if the robot didn't know where it was. It had to traverse the halls, and it would have to get back to a charger eventually. But a human could help with that. Maybe we could make Kuri ✨wander✨?

[Nick Walker @nickwalker_us]

We wanted to study some interactions with a Kuri (discontinued sub-$1000 home robot) out in our building, but its lidar was too weak for conventional autonomous navigation, and its computer was no match for visual navigation methods. Teleoperating for days was out of the question

[Nick Walker @nickwalker_us]

Some of the longest running research deployments use expensive, heavy duty platforms to navigate autonomously. This takes a lot of engineering effort, and it's still not completely robust. This BWIbot was defeated by a shiny bench.

[Nick Walker @nickwalker_us]

Work with @mavrojean @siddhss5 @mayacakmak.
Hire Chris, he'll be a good advisor.

[Nick Walker @nickwalker_us]

Good morning #CoRL2021! I'll be sharing a recipe for making your robots sensitive to ✨behavioral attributions✨ during the poster session later today. Why? 🧵 https://t.co/2UbT2abqTZ

[Nick Walker @nickwalker_us]

I'm not there in person, which is a bummer because way back I bought this shirt with the sole purpose of wearing it at a conference whenever we published this. Please imagine me walking around the venue in this shirt. Thanks.