[Concept] FOSS/OSH home assistance robot (Labrador Systems open equivalent) #43

New issue

Open

opened 2026-05-31 09:09:29 -07:00 by pyr0ball · 0 comments

pyr0ball commented

2026-05-31 09:09:29 -07:00

Owner

Concept

Build a privacy-first, FOSS/OSH autonomous home assistance robot filling the same niche as the Labrador Systems Retriever: an autonomous mobile platform that fetches, carries, and delivers items around the home to support people with mobility limitations, chronic pain, or other physical access needs.

Labrador Systems' product is backed by Amazon's Alexa Fund, uses Alexa for voice control, and is proprietary. The CF equivalent replaces every closed-source/cloud dependency with FOSS or OSH alternatives.

Target users

Primary: people with mobility limitations, chronic pain, fatigue disorders, post-surgical recovery, or other conditions that make carrying and fetching physically costly. Strong overlap with CF's ND/adaptive needs market — executive function load reduction, independence, reduced reliance on caregivers.

Core functions

Autonomous navigation (room-to-room, through doorways)
Item fetch and carry (tray/shelf platform)
Voice command interface (local STT/TTS, no cloud)
Scheduled and app-triggered delivery
Self-docking/charging

FOSS/OSH stack candidates

Layer	Labrador (proprietary)	CF equivalent
Voice control	Amazon Alexa	Whisper STT + local LLM + Piper TTS
Navigation	Proprietary	ROS2 Nav2 + SLAM
Platform hardware	Proprietary cart	OSH wheeled platform (SCUTTLE, modified
Home integration	Proprietary app	Home Assistant (HASS)
Visual understanding	Proprietary	Fine-tuned VLM on SynData egocentric clips
Task planning	Proprietary	Local LLM mid-trained on SWE-ZERO corpus

Training data connections

PsiBotAI/SynData (roadmap#42): egocentric manipulation captions for visual grounding
AlienKevin/SWE-ZERO-12M-trajectories (roadmap#41): agentic tool-use priors for task planning LLM

CF design principles

Local inference first: all voice/vision/planning runs on-device, no cloud required
No PII logging
Open hardware wherever possible; where proprietary parts are unavoidable, document substitutions
Accessibility is the product, not a feature

Product naming

Pending. Avian convention applies. Candidates: swift (agile, always working), nuthatch (climbs/navigates, industrious), dipper (already in backlog queue). Open to input.

Next steps

Survey existing OSH mobile platform options (SCUTTLE, others)
Map minimum viable hardware BOM
Prototype ROS2 Nav2 on existing Heimdall/Navi hardware before committing to custom platform
Define voice command intent schema and tool-call interface
Assess SynData viability for visual grounding (see roadmap#42)

## Concept Build a privacy-first, FOSS/OSH autonomous home assistance robot filling the same niche as the Labrador Systems Retriever: an autonomous mobile platform that fetches, carries, and delivers items around the home to support people with mobility limitations, chronic pain, or other physical access needs. Labrador Systems' product is backed by Amazon's Alexa Fund, uses Alexa for voice control, and is proprietary. The CF equivalent replaces every closed-source/cloud dependency with FOSS or OSH alternatives. ## Target users Primary: people with mobility limitations, chronic pain, fatigue disorders, post-surgical recovery, or other conditions that make carrying and fetching physically costly. Strong overlap with CF's ND/adaptive needs market — executive function load reduction, independence, reduced reliance on caregivers. ## Core functions - Autonomous navigation (room-to-room, through doorways) - Item fetch and carry (tray/shelf platform) - Voice command interface (local STT/TTS, no cloud) - Scheduled and app-triggered delivery - Self-docking/charging ## FOSS/OSH stack candidates | Layer | Labrador (proprietary) | CF equivalent | |---|---|---| | Voice control | Amazon Alexa | Whisper STT + local LLM + Piper TTS | | Navigation | Proprietary | ROS2 Nav2 + SLAM | | Platform hardware | Proprietary cart | OSH wheeled platform (SCUTTLE, modified | | Home integration | Proprietary app | Home Assistant (HASS) | | Visual understanding | Proprietary | Fine-tuned VLM on SynData egocentric clips | | Task planning | Proprietary | Local LLM mid-trained on SWE-ZERO corpus | ## Training data connections - `PsiBotAI/SynData` (roadmap#42): egocentric manipulation captions for visual grounding - `AlienKevin/SWE-ZERO-12M-trajectories` (roadmap#41): agentic tool-use priors for task planning LLM ## CF design principles - Local inference first: all voice/vision/planning runs on-device, no cloud required - No PII logging - Open hardware wherever possible; where proprietary parts are unavoidable, document substitutions - Accessibility is the product, not a feature ## Product naming Pending. Avian convention applies. Candidates: swift (agile, always working), nuthatch (climbs/navigates, industrious), dipper (already in backlog queue). Open to input. ## Next steps - [ ] Survey existing OSH mobile platform options (SCUTTLE, others) - [ ] Map minimum viable hardware BOM - [ ] Prototype ROS2 Nav2 on existing Heimdall/Navi hardware before committing to custom platform - [ ] Define voice command intent schema and tool-call interface - [ ] Assess SynData viability for visual grounding (see roadmap#42)