VLA GalleryAll modelsGallery

VLA Gallery / All models / Industry models

Gemini Robotics-ER 1.5

Google DeepMind · 2025-09

기타Industry modelsNot disclosedOther

A high-level 'brain' that orchestrates robot activity, issuing step-by-step natural-language instructions to GR 1.5.

View architecture diagram in the galleryPaper ↗

Paper

Gemini Robotics 1.5: Pushing the Frontier of Generalist Robots with Advanced Embodied Reasoning, Thinking, and Motion TransferarXiv 2510.03342

Overview

Category
Industry models
Type
기타
Released
2025-09
Organization
Google DeepMind
Open weights
No
Parameters
Not disclosed

Architecture

Backbone LLM
Based on Gemini (size not disclosed)
Vision encoder
—
Action head
Points, bboxes, trajectories, and tool calls instead of actions (Other)
Chunk
Not disclosed
Control rate
Not disclosed
Architecture (original summary)
Gemini-based thinking VLM (retains all Gemini capabilities); no direct action output (2D points · bboxes · trajectories, tool calls)

Training

Pretraining accelerators
TPU v4, v5p, v6e (chip count not disclosed; model card shared with GR 1.5)
Pretraining duration
Not disclosed
Pretraining steps & batch
Not disclosed
GPU/TPU-hours
Not disclosed
Fine-tuning
Not disclosed
Data type
Image·text·robot sensor·action data (model card description shared with GR 1.5)
Data scale
Not disclosed
Main datasets
Not disclosed
Tasks & evaluation
15 academic benchmarks (ERQA, Point-Bench, BLINK, CV-Bench, RoboSpatial, PointArena, Where2Place, RefSpatial, etc.); 8 long-horizon ALOHA and bi-arm Franka tasks
Robot
ALOHA, bi-arm Franka (orchestrates GR 1.5 on long-horizon tasks)

Inference

Inference device & speed
Gemini API 'gemini-robotics-er-1.5-preview' (Google AI Studio, 2025-09-25); latency-accuracy trade-off via thinking budget; latency figures not disclosed
Inference memory
Not disclosed
Weights bf16 (calc.)
Not disclosed
Where it runs
Cloud (Gemini API)

Reinforcement learning & base

Reinforcement learning
Not disclosed · No mention of RL in technical report
RL evidence
https://arxiv.org/html/2510.03342
Base control
Fixed · No action output (as an orchestrator, outputs only text, coordinates, and tool calls)

Notes

Key features
A high-level 'brain' that orchestrates robot activity, issuing step-by-step natural-language instructions to GR 1.5. Supports Google Search·VLA·user function calls and thinking budget control.
Notes
A VLM (ER) that does not output actions directly, so type is Other. Blog 2025-09-25, arXiv v1 2025-10-02. The model card Training Data row describes ER 1.5 and GR 1.5 together. 2026 API docs refer to ER 1.6/ER 2.
Verification
Confirmed

Other models in this category