DAG-Based Planning (LLM + Verifier)
LLM Planner (Qwen3-Next-80B)
Intent -> JSON plan, retry + temperature escalation
DAG Parser + Schema Validation
Dependency graph, sequential/DAG topology
Symbolic Verifier (6 rules)
Actions, cycles, orphans, battery, safety, depth
Dynamic Replanning
SCAN_AREA replan trigger
Keywords or interesting=true -> cancel + refine + re-execute
4 passing integration tests
Executor + server dynamic replan flow validated
State Machine (7 states)
IDLE -> LISTENING -> PROCESSING
-> AWAITING_CONFIRM -> EXECUTING
-> SPEAKING -> IDLE
Robot Actions (8 total)
CHECK_BATTERY - edge proxy telemetry
NAVIGATE - NavigateToPose (partial waypoint)
SCAN_AREA - frame -> VLM -> analysis + replan
IDENTIFY_PERSON - frame -> FR -> identity
VERIFY_OBJECT - frame -> VLM -> verification
SPEAK / WAIT / ALERT_OPERATOR
Test Coverage: 14 test files
State machine, DAG parser, symbolic verifier
FR/TTS/VLM clients, FR temporal analyzer
Dashboard confirmation, dynamic replanning
Services Integrated
FR Client (WS to :42067), TTS Client (HTTP to :8200)
VLM Client (modelapi chat), Edge Proxy (WS to :8080)
FR Temporal Analyzer (10s window dedup)