SOMA & ANIMA
Latest robot

SOMA Arm

The latest robot on display is SOMA Arm. It is the first public robot used to validate ANIMA, real robot execution, and the broader embodied AI direction.

A tabletop language-driven robot arm project used to validate ANIMA's cognition loop, visual grounding, skill execution, and failure recovery.

Explore SOMA ArmRead the ANIMA framework
SOMA Arm
Current public track

It is not the final product. It is the benchmark and manipulation capability layer on the path to home robotics.

First make language understanding, skill dispatch, visual validation, and real robot execution reliable. Then carry that stack into richer domestic environments and longer-horizon tasks.

SOMA Family

The currently public SOMA family robots

Two product lines in the current public SOMA family: SOMA Arm (tabletop chess manipulation) and SOMA Care (simulated-ward care). Both share the same ANIMA cognition framework.

SOMA Arm
The real workstation, arm, and chessboard form the clearest system carrier right now.
In DevelopmentV1.01·Latest: camera stabilization, intrinsic calibration, eye-to-hand calibration, world grounding

SOMA Arm

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.

Open profile ->
SOMA Care
Prototype view of the Stretch RE3 in the simulated ward — bed, nightstand, bedside table, and care robot in one frame.
In DevelopmentV0.4·Latest: PBR ward assets, contact-rich skills, 8–10 success paths + 3–5 failure paths

SOMA Care

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.

Open profile ->
Explore SOMA ArmExplore SOMA Care
ANIMA

The ANIMA cognition framework

This section mainly introduces ANIMA. It is the shared cognition framework responsible for language understanding, task structure, skill dispatch, and result validation across the SOMA robot line.

Cognition framework · v0.1.0

ANIMA

A
utonomous
N
atural-language
I
nstruction
M
apping
A
rchitecture

Turn upstream signal into verifiable, dispatchable, recoverable robot action.

Signal inputs
textvoiceBCIvision
Action guarantees
VerifiableDispatchableRecoverable
Ships as a standalone pip-installable Python library · jeffliulab/anima
Open the ANIMA framework page
Core structure
L5
Assessment

Five-factor self-assessment (ITA/MQA/SQA/GOA/PEA), event-triggered — not continuous logging

L4
Adapter

LLMToolCaller + EmbodiedAdapter protocols decouple the embodiment

L3
Skill

py_trees behavior-tree runtime · Function-Calling + Affordance Scoring

L2
Planner

TaskSpec orchestrated into an executable behavior tree

Test-and-Check · 6 gates
JSONIntentSkillParamsSafetyPrecond.
L1
Parser

LLM-as-Parser — forced tool-calling emits TaskSpec JSON

L0
Signal

Upstream signal entry — text / voice / BCI / vision

Five-factor self-assessment (event-triggered)
ITAIntent TruthMQAMission QualitySQASkill QualityGOAGoal OutcomePEAPersistent Effect
Development Progress

Latest development progress

The homepage now shows only the latest progress updates, while the full version-based development log remains on the Devlog page.

Latest timeline
2026-04-21ANIMA

ANIMA v0.1.0 — first reference implementation shipped

The L0–L5 six-layer stack, six-gate Test-and-Check, and five-factor self-assessment (ITA / MQA / SQA / GOA / PEA) now ship together as a standalone pip-installable Python library at jeffliulab/anima.

2026-04-21SOMA Care

SOMA Care v0.4 kicked off — second product line plugs into ANIMA

Simulated-ward ADL intent-to-action loop; six intents running end-to-end; parallel to soma-arm and sharing the same ANIMA cognition framework.

2026-04-14 ~SOMA Arm

V1.01 — Perception Foundation

Camera parameter lock, intrinsic calibration, eye-to-hand calibration, world coordinate grounding, workspace reachability validation.

SOMA Arm devlogSOMA Care devlogANIMA devlog
Open-source Repositories

Open-source GitHub repositories

The public repositories today: the two SOMA product lines (SOMA Arm and SOMA Care) plus their shared cognition framework ANIMA.

repo

soma-arm GitHub repo

The most concrete public project today and the best first portfolio link.

Open the GitHub repo ->
repo

soma-care GitHub repo

The care product line of SOMA — a second reference implementation for the intent-to-action loop, running in a simulated ward.

Open the GitHub repo ->
repo

ANIMA GitHub repo

The framework entry point for engineers who want the cognition-layer story and public structure.

Open the GitHub repo ->