Case study 05 / University project · Embodied AI
ROS2 Robot Vision & Navigation
From a classifier to a physical control loop.
MobileNetV3-Large on a physical robot, connected to a four-state navigation loop.
01 / Context
Problem
A model running on a robot has to connect perception with behaviour. This project pairs a MobileNetV3-Large classifier with a physical ROS2 robot and an explicit navigation loop.
The loop has four states: SEARCH, ALIGN, APPROACH and ARRIVED.
02 / Ownership
My Role
Worked on the MobileNetV3-Large classifier running on a physical robot and the four-state navigation loop in ROS2.
03 / System design
Architecture
- 01Robot perception
Visual input on the physical robot
- 02Classification
MobileNetV3-Large
- 03State control
SEARCH → ALIGN → APPROACH → ARRIVED
- 04Robot behaviour
Navigation on physical hardware
04 / Engineering decisions
Engineering Decisions
Making the control structure explicit.
The four named states make the intended progression from search to arrival visible in the architecture.
A recording of the physical robot and documented recovery behaviour would make the strongest next evidence for this case study. Neither an unseen demonstration nor unreported success rate is claimed here.
05 / Evidence
Results
The supplied evidence establishes the physical robot, classifier and state loop. Navigation success rate, inference latency and hardware details have not been supplied.
06 / Tools & capabilities
Stack
07 / Explore further
Links
The exact project repository will be added when available.