Quest teleop Single Module

[leshy]

reviewing #1215 and looks ok, just there is a cleaner arch approach.

For now we have in one terminal a parallel TS system that collects raw pose/buttons data and publishes to LCM
quest -> typescript deno server -> LCM
these are not dimos modules, this is typescript publishing to LCM directly

And another terminal runs actual dimos
LCM -> questTeleopModule(s) -> Poses, Twists, etc

Result is that we have two parallel systems (TS and dimos), hard dependancy on LCM as transport between them, with blueprints not understanding topics used.

Ideally as a dimos user I want just
quest -> questTeletopModule -> Poses, Twists, etc

And I expect questTeletopModule to setup web server and whatever is required.

This is easily achiveable. Instead of sending fully encoded LCM packets (that include topics encoding and are passed directly to UDP) you just encode actual data (without a topic)

can read first of all about how LCM provides us with data encoding methods on all msgs https://github.com/dimensionalOS/dimos/blob/main/docs/concepts/lcm.md

from https://jsr.io/@dimos/msgs

// Construction with partial init
const pose = new geometry_msgs.Pose({
  position: new geometry_msgs.Point({ x: 1, y: 2, z: 3 }),
  orientation: new geometry_msgs.Quaternion({ w: 1 }),
});

// Encode to bytes (includes 8-byte hash prefix)
const bytes: Uint8Array = pose.encode();

send those bytes off

on the python side you'd start a simple webserver with websocket connection.
then when receive a msg,

Then you have 2 options

publish those bytes directly to a standard transport without decoding

(transport needs to accept LCM encoding, it uses from dimos.protocol.pubsub.encoders import LCMEncoderMixin)

self.left_pose.publish(received_bytes) (#1223 needs merge for this first)

This way you can have standard transports on your module, blueprints understand it etc, and it's very efficient since you are not decoding messages coming from quest, just passing them

decode & do whatever

this is actually better imo

left_pose = PoseStamped.lcm_decode(bytes)

and do whatever you wish with the pose object, convert to twist, do those follow up control transformations, publish when ready

left_pose = PoseStamped.lcm_decode(those_bytes)
self.left_twist.publish(left_pose.to_twist())

This way you don't have 2 dimos modules for quest (receiver/server + conversions for control like twist etc) but a single module