Hi everyone,
I'm Peter, an architect in New York. I've spent my career at the intersection of architecture and robotics (Toyota Woven City, 3D Printed Construction, etc.). Lately I keep hearing the same thing from robot teams: the robot drives fine, and the building is the hard part. Every new site has a different elevator company, a different door and badge system and its own IT review, so the robot waits in the lobby while someone builds one more integration.
So I'm building **plynth**: one API between robots and a building's elevators and doors. Your robot asks to ride or to go through; the building's gateway calls the car or opens the door, or says no with a reason, under rules the owner sets (which robots, where, when). The idea is that you integrate once, and each new building is our work, not yours.
**[Try the sandbox](https://sandbox.plynth.build/docs)** · **[plynth.build](https://plynth.build)**
You can try it now against a test building. One click on the docs page gives you a key for a day, no sign-up, and the Python kit has nothing to install. With the key loaded, a door passage is:
```python
import os, uuid
from plynth.client import RobotClient
robot = RobotClient(os.environ["GATEWAY_URL"], os.environ["GATEWAY_ROBOT_ID"], os.environ["GATEWAY_ROBOT_SECRET"])
trip = str(uuid.uuid4()) # a new ID for each trip; sending it again is safe
print(robot.request(trip, "door")) # state OPEN_COMMANDED: go through
print(robot.report(trip, "passed")) # state PASSED, done: true, final: true
```
`python3 -m plynth.hello` checks your key, and `python3 -m plynth.conformance` rides the elevator, goes through the doors and reports what passed. Underneath is a plain HTTPS API: requests signed with HMAC-SHA256, a stream for following a ride, and a reason code on every refusal. The docs have the signing recipe in shell, Python, JavaScript and C++.
The test building has an elevator bank, a door and a badge door, plus doors and elevators that fail on purpose: a jammed door, a slow car, fire service recall mid-ride, a car that drops off Wi-Fi. The point is to write the handler for each one before a site visit, not during it.
**For ROS 2** (Jazzy):
- `plynth_ros_example`: one node that goes through a door, then rides an elevator.
- `plynth_rmf_adapter`: an Open-RMF lift and door adapter. Each lift session and door request becomes a passage the building approves or refuses.
- `plynth_node`: the building's holds on `/plynth/stop`, as a twist_mux lock. This one needs a company key.
The kit, the conformance check and the ROS 2 packages are Apache-2.0, on GitHub at `plynthbuild/plynth-robot-kit`. plynth itself is a commercial service; the sandbox is free. There's no live building yet. I'm looking for a few robot teams to try it, tell me what's wrong, and go into the first buildings with us.
If you've run robots on real elevators, I'd especially like to hear what broke that a simulator wouldn't show, and what building owners asked for before they said yes.
Peter
What’s wrong is the first part: the hard part (integrating with actual elevators) isn’t done yet. All of this is would be nice. Integrating with elevators is not only time consuming and costly. There are also no guarantees up-front that it will be successful for each new building you want to add your robots to. We learned this the hard way at Savioke (Relay).
You are working on the easy part. People will be much more interested in this once you show some success with the hard one.
Thanks, I really appreciate the feedback(and sorry for the slow reply). Fair point. The robot-side API is the easy half, and nothing proves this until a real robot rides a real elevator through plynth in a live building. That’s the next milestone. The elevator manufacturer robot APIs work against their virtual building, but that isn’t the same thing.
The “no guarantee per building” part is what I keep coming back to, and from what I can find it’s still true. Maker APIs only reach cars the maker itself maintains, and I found that most recently re-controlled cars have independent controllers.
When a building didn’t work out at Savioke/Relay, what usually decided it? Would love to get your thoughts on strategies for integration.