ANIMA's L0–L5 stack translates upstream signals (text, voice, BCI, vision) into verifiable robot actions layer by layer, each layer carrying a clear input/output contract and a named failure path.
Upstream signal gateway: turns BCI / ASR / vision / text into an intent token + confidence + drift_score. Drift propagates downstream as an uncertainty handle.
The LLM is locked to forced tool-calling, compressing natural-language instructions into a structured TaskSpec JSON. It never emits motor commands — only structure.
TaskSpec → py_trees behavior tree. Conditional branching, retry, and fallback live in the tree itself — no ad-hoc state machines.
Function-Calling + Affordance Scoring selects skills (more accurate than RAG when the skill set is under 100). Preconditions and expected effects are declared in the skill registry.
Device-agnostic actuation. The same L1–L3 drives a manipulator, a mobile base, a wheelchair, or a future humanoid — because each L4 implements the EmbodiedAdapter protocol.
The five factors (ITA / MQA / SQA / GOA / PEA) each evaluate at Pre / Runtime / Post stages, producing a success probability with calibrated confidence.
The most concrete reference implementation today
A tabletop language-driven robot arm project used to validate ANIMA's cognition loop, visual grounding, skill execution, and failure recovery.
It is not the final product. It is the benchmark and manipulation capability layer on the path to home robotics.
Medical-care intent-to-action simulation
The medical-care product line of the SOMA family. Powered by the ANIMA cognition framework, it turns intent (text, voice, and future BCI signals) into auditable robot actions inside a simulated hospital ward. v0.4 focuses on showcasing the full intent-to-care loop via MuJoCo simulation and offline video.
The care product line of the SOMA family, running in parallel with soma-arm and sharing the ANIMA cognition framework.