Skip to content

README.md

Latest commit

 

History

History
59 lines (35 loc) · 3.78 KB

README.md

File metadata and controls

59 lines (35 loc) · 3.78 KB
Raw

SynROV AiBot — Software Version 1

AiBot provides the Python intelligence/runtime layer for SynROV. It includes the runtime (runtime.py), desktop Command Center (app.py), physical-camera support (camera_devices.py), robot-specific command/intention catalog (robot_catalog.py), perception, safety, orchestration, voice input, and protocol handling.

This Python package is part of SynROV software version 1. Shared application messages use softwareVersion: 1, while schema, variable, file, and directory names remain stable.

Run

python -m pip install -r requirements.txt
python synrov_aibot.py

Headless:

python -m synrov_aibot --headless --uri ws://127.0.0.1:9000/ --loop
python -m synrov_aibot --headless --uri ws://127.0.0.1:9000/ --loop --camera auto

Communication

AiBot uses the same canonical SynROV envelope as the browser and ROS:

  • 9000: control/state;
  • 9001: Processing 3D perception (perception.subscription / perception.frame);
  • 9002: Vehicle/Drone camera (robot_camera.control / robot_camera.frame).

See ../../docs/COMMUNICATION_PROTOCOL.md for the canonical contract.

Language packages and voice

AiBot loads external JSON packages from synrov_aibot/languages/. English is the fallback package; AiBot does not expose an independent language selector and always follows the package code published by Processing. The selected package updates the Command Center, robot command/intention display text, and the speech-recognition locale without restarting the microphone path.

Static Command Center widgets are authored from canonical English keys and translated through the package table; no Portuguese/English bilingual literal is used as the rendered source. The language-consistency tests also require exact key parity across all eight AiBot packages, preventing a missing translation from silently producing a mixed-language screen.

Audio capture uses sounddevice with 16 kHz mono PCM, ambient-noise calibration, and local phrase segmentation. SpeechRecognition is used for transcription. See ../../docs/LANGUAGE_PACKS.md for the package schema.

Robot-specific intelligence

robot_catalog.py does not define one global command set. Manipulator, Vehicle, and Drone have their own direct commands and missions. The same catalog displayed in the Intentions tab is used by the command resolver.

The Operation tab exposes adaptive AI, music/rhythm, vision/object handling, and webcam controls. Learning protection, shadow validation, arbitration, sensor fusion, safety, orchestration, and performance decisions remain internal to the runtime. Bounded mission-strategy learning and teachable spoken-phrase aliases run independently for Manipulator, Vehicle, and Drone and persist per robot. Long-context episodic memory is also isolated per robot; only music/rhythm actuation remains restricted to Manipulator.

Heading/compass is part of the sensor context for all three robots. Vehicle and Drone use camera pan -60..+60° and tilt -20..+20°.

Learned runtime data is persisted under synrov_aibot/runtime_data/ and is not distributed as source code.

Tests

python -m compileall synrov_aibot
python -m unittest discover -s tests -v

Bounded mission-strategy learning

AiBot does not train arbitrary motor outputs. AdaptiveMissionPolicy selects only pre-approved safe variants for selected search/inspection missions, observes mission outcomes, and stores aggregate rewards in runtime_data/models/<robot>/<robot>_policy.pkl plus JSONL experience records in runtime_data/datasets/<robot>_training.jsonl. Strategy selection is persistent across launches. Direct commands, emergency actions, pick/place actions, and origin-navigation safety logic are not raw-exploration targets. Processing and firmware safety remain authoritative.