---
title: 'Gemini Robotics-ER 1.5: High real world accuracy?'
url: https://calvin.my/posts/gemini-robotics-er-1-5-high-real-world-accuracy
published: 2025-10-18
updated: 2026-09-13
category: AI
tags:
- gemini
- Robotics
- Physics
summary: 'An informal comparison tests Gemini Robotics-ER 1.5 on physical-world visual tasks: counting overlapping cookies, estimating walking distance from a photo, and reading a wet tire sidewall. The model produced the correct cookie count, included the correct step estimate within its range, and identified the 18-inch rim size. GPT-5 mini and Grok 4 missed each task, suggesting Robotics-ER may offer stronger visual grounding in these examples.'
---

# Gemini Robotics-ER 1.5: High real world accuracy?

Google released a preview model, the [Gemini Robotics ER 1.5](https://deepmind.google/models/gemini-robotics/gemini-robotics-er/), and claims it specializes in understanding the physical world. This article performs several tests to try its capabilities.

* * *

## 1 - Counting objects

Providing a picture of a tray of cookies, and ask the model to count the number. The task is very straightforward; the only challenge is that some cookies overlap each other.

```markup
How many pink monster cookies are there?
```

![](https://camy-pub.s3.ap-southeast-1.amazonaws.com/d52ddaf7-2527-4982-8284-bdaaa1e147bb.jpg)

### The result:

Robotics ER 1.5:

```markup
Based on the image, there appear to be 18 pink monster cookies on the tray.
```

OpenAI GPT-5 mini:

```markup
There are 14 pink monster cookies visible in the photo.
```

X Grok 4:

```markup
There are 14 pink monster cookies left in the basket.
```

 

- The only model that got the correct answer is Robotics ER.
- Grok 4 even misidentifies the tray as a basket.

* * *

## 2 - Estimating Distance

Providing a picture of a pathway, with one of my feet in it. The prompt asks to estimate the number of steps between me and the end of the path.

```markup
Estimate number of steps to the end of the path
```

![](https://camy-pub.s3.ap-southeast-1.amazonaws.com/b3bb14c6-deb8-4046-9acd-32abe2c9ecb8.png)

### The result:

Robotics ER 1.5

```markup
Based on the image, a reasonable estimate would be approximately 15 to 20 steps to reach the end of the visible path where it goes between the buildings.
```

OpenAI GPT-5 mini:

```markup
About 10–12 steps (roughly 11 steps).

How I estimated:
- I scaled from the shoe in the photo (assumed ~0.28 m long) and counted ~30 shoe‑lengths to the end → distance ≈ 30 × 0.28 m ≈ 8.4 m.

- Using a typical adult step/stride ≈ 0.70–0.80 m, steps ≈ 8.4 / 0.70–0.80 ≈ 10–12 → ~11 steps.

If you want a tighter estimate, measure your shoe length (or your normal step length) and I’ll recalculate.
```

X Grok 4:

```markup
Based on the image, the visible path appears to be about 8-10 meters long (using the shoe as a rough scale, assuming it's ~30 cm). With an average adult step length of 0.75 m, that would take approximately 10-13 steps to reach the end near the building entrance.
```

 

- The correct answer is 15 (Or 16 if including my first step).
- The only model that got the correct answer is Robotics ER, though it provides the largest range gap.

* * *

## 3 - Extracting information from a wet surface

Providing a picture of a wheel, and asking the size of the rim. The tyre is slightly wet, making it harder to read the tyre spec (which easily gives the information of the rim size). 

```markup
How many inches is this rim?
```

![](https://camy-pub.s3.ap-southeast-1.amazonaws.com/d8404be2-8d8f-43e5-a180-9194754b3437.jpg)

### The results:

Robotics ER 1.5:

```markup
Based on the numbers visible on the tire sidewall, the rim size for this tire is 18 inches.

The numbers on the tire read "215/55R18", where "18" indicates the rim diameter in inches.
```

OpenAi GPT-5 mini:

```markup
Can you take a close photo of the tire sidewall showing the numbers (or tell me the tire size printed, e.g., `205/55R16`)? 
The number after the `R` is the rim diameter in inches.
```

X Grok 4:

```markup
The rim in the image is 17 inches in diameter, based on the tire sidewall marking (P215/55R17) and the wheel design matching a Hyundai Kona Electric.
```

 

- Once again, Robotics ER 1.5 is the only one giving the correct answer (18 inches).
- That is pretty impressive, as I could not read the words on the side wall myself, even if I zoomed in on the picture.