Architecture

VLS

ExperimentalPublished: 28 July 2026Updated: 28 July 2026Published

Key innovation

Separates 'what' from 'how': a vision-language model sets the task, while execution is carried out by pre-validated, repeatable skills that are retrieved and composed, instead of monolithic end-to-end action prediction as in VLA models.

How it works

1) A vision-language model interprets the instruction and the scene, setting the task goal (WHAT to do). 2) A planner/orchestrator decomposes the task into a sequence of skills and retrieves the appropriate validated skills from the library. 3) Each skill performs a specific, repeatable action (HOW), often pre-tested in simulation and reality. 4) Skill outcomes are monitored (measurability), and the orchestrator corrects/continues the plan until the task is complete.

Problem solved

End-to-end VLA models can be unpredictable, hard to validate and error-prone out of distribution. VLS increases reliability, repeatability and testability by basing execution on pre-validated skills.

Components

Vision-language module (VLM)Task understanding

Interprets the scene and language instruction, setting the task goal (WHAT to do).

Official

Skill libraryExecution (HOW)

A collection of pre-validated, repeatable skills performing specific actions (HOW).

Official

Orchestrator / plannerComposition and control

Decomposes the task into skills, retrieves them from the library, composes them into a sequence and monitors execution.

Official

Implementation

Reference implementations

Holiday Robotics — FRIDAY (VLS)

Holiday Robotics

Official

Implementation pitfalls

Skill-library gaps (long tail)High

Tasks needing skills outside the library are infeasible; covering the long tail is hard.

Fix:Continuously expand and learn new skills, fall back to learning from demonstration.

Orchestration/composition errorsMedium

Wrong choice or ordering of skills breaks the task despite correct individual skills.

Fix:A better planner, in-loop monitoring and correction, pre/post-condition checks.

Skill maintenance and validationMedium

Each skill requires validation (sim + real) and maintenance, which scales expensively.

Fix:Automated skill testing, standardized interfaces, reuse.

Evolution

2023

Dominance of end-to-end VLA

Vision-Language-Action models (e.g. RT-2) establish the monolithic, end-to-end approach to robot control that VLS differentiates from.

2026

Holiday Robotics introduces VLS (FRIDAY)

Inflection point

Holiday Robotics promotes the Vision-Language-Skill architecture for the FRIDAY humanoid as a modular, validated alternative to end-to-end VLA (a $105M Series A).

Hyperparameters (configurable axes)

Skill granularityHigh

How fine-grained/complex individual skills are.

Skill-library sizeCritical

Number and coverage of validated skills.

Orchestration methodHigh

How the task is decomposed and composed from skills (planning, LLM, rules).

Computational complexity

Time complexity: Zalezna od VLM (planowanie) + wybor/wykonanie umiejetnosci. Space complexity: O(VLM + biblioteka umiejetnosci).

Compute bottleneck

Skill-library coverage and quality

VLS effectiveness depends on task coverage by validated skills; library gaps (the long tail) and orchestration errors limit flexibility relative to end-to-end VLA. Validating skills (sim + real) is costly.

Depends on

Kompletnosc biblioteki umiejetnosciJakosc orkiestracji