# Can a home robot patrol autonomously with only a monocular camera, no odometry, and cloud computing? — retained record

> Retained reddit record from reddit:robotics.

Canonical: https://highsignal.app/data/records/8843812d262d20fb

[← sources](https://highsignal.app/data)


source: [reddit](https://highsignal.app/data/reddit) (reddit:robotics)  published 2026-09-26 03:24 ingested 2026-09-27 02:39

[original source ↗](https://www.reddit.com/r/robotics/comments/1wqfzit/can_a_home_robot_patrol_autonomously_with_only_a/)

## retained text



```
Hi everyone,

I’m evaluating whether a low-cost wheeled robot can perform autonomous patrols inside a home. The hardware is fixed, and I’d appreciate a realistic assessment of what is achievable.

**Hardware and computing constraints**
- One monocular RGB camera, with no depth output.
- Camera stream: 15 FPS.
- No IMU, wheel encoders, or LiDAR.
- No onboard computing available for SLAM or navigation; processing must run in the cloud.
- Images are sent to the cloud, and movement commands are sent back to the robot.
- Adding or replacing sensors is not an option.

**What I want it to do**

The goal is to visit predefined locations in different rooms and capture images for monitoring, then repeat the route on later patrols. This is primarily a navigation question, not an image-understanding question.

For an initial prototype, I’m open to manually driving the robot around to build a map first. Fully autonomous exploration is not required.

**My questions**

1. Is monocular visual SLAM a practical basis for this task under these constraints, or would another approach, such as visual teach-and-repeat, be more suitable?
2. Without an IMU or wheel odometry, how would you handle scale ambiguity, tracking loss, and relocalization after a restart?
3. Could this setup support useful obstacle avoidance, including people or objects that were not present during mapping?
4. How much would cloud latency and jitter affect navigation? What end-to-end latency and effective frame rate would you need to measure before deciding whether this is viable?
5. With the hardware unchanged, what is a realistic first milestone: repeating a taught route in controlled conditions, navigating between mapped waypoints, or something more limited?

I’m particularly interested in firsthand experience or open-source implementations tested on physical robots. Examples that rely on additional sensors may not transfer to this setup, so please mention the sensors used.

If reliable unattended patrol is unrealistic, I’d also appreciate an explanation of the main limiting factor and what restrictions on the environment or task could make a prototype workable.

Thanks!
```



document fetched 2026-09-27 00:35 · canonical https://reddit.com/r/robotics/comments/1wqfzit/can_a_home_robot_patrol_autonomously_with_only_a
