Backlog: YOLO26 for Nuthatch robot vision (object detection + pose on Jetson) #52

New issue

Open

opened 2026-06-23 09:30:12 -07:00 by pyr0ball · 0 comments

pyr0ball commented

2026-06-23 09:30:12 -07:00

Owner

Source

Paper: https://arxiv.org/abs/2606.03748 — Ultralytics YOLO26 (June 2026)
Code/weights: https://github.com/ultralytics/ultralytics

What it is

Unified real-time vision model: detection, segmentation, pose estimation, oriented
detection, classification in one pipeline. NMS-free. TensorRT export path for
NVIDIA hardware (1.7–11.8ms on T4).

Nuthatch use cases (when repo is created)

Object detection: Identify obstacles, furniture, humans in navigation path.
Pose estimation: Track human body pose for intent/proximity estimation —
"is the person reaching for something?" — useful for assistive robot behaviors.
Person detection: Safety stop — halt motion if human enters proximity zone.
Jetson hardware fit: TensorRT export optimized for NVIDIA Jetson. Nuthatch's
on-hand Jetson hardware should get near-paper-level latency.
Shared foundation: cf-vision module (roadmap#51) will abstract this — Nuthatch
should consume cf-vision rather than integrating YOLO26 directly.

License situation

AGPL-3.0 for open source; commercial license for distribution. For Nuthatch (deployed
on user's own hardware, not SaaS), AGPL-3.0 is acceptable in the free tier.

roadmap#51: cf-vision module (the right abstraction layer for Nuthatch to use)
cf-vision repo: YOLO26 integration ticket
Nuthatch hardware: Jetson, wheelchair base, Xtion, webcams

## Source Paper: https://arxiv.org/abs/2606.03748 — Ultralytics YOLO26 (June 2026) Code/weights: https://github.com/ultralytics/ultralytics ## What it is Unified real-time vision model: detection, segmentation, pose estimation, oriented detection, classification in one pipeline. NMS-free. TensorRT export path for NVIDIA hardware (1.7–11.8ms on T4). ## Nuthatch use cases (when repo is created) - **Object detection:** Identify obstacles, furniture, humans in navigation path. - **Pose estimation:** Track human body pose for intent/proximity estimation — "is the person reaching for something?" — useful for assistive robot behaviors. - **Person detection:** Safety stop — halt motion if human enters proximity zone. - **Jetson hardware fit:** TensorRT export optimized for NVIDIA Jetson. Nuthatch's on-hand Jetson hardware should get near-paper-level latency. - **Shared foundation:** cf-vision module (roadmap#51) will abstract this — Nuthatch should consume cf-vision rather than integrating YOLO26 directly. ## License situation AGPL-3.0 for open source; commercial license for distribution. For Nuthatch (deployed on user's own hardware, not SaaS), AGPL-3.0 is acceptable in the free tier. ## Related - roadmap#51: cf-vision module (the right abstraction layer for Nuthatch to use) - cf-vision repo: YOLO26 integration ticket - Nuthatch hardware: Jetson, wheelchair base, Xtion, webcams