ARGOS-MUKHTAR is a research-stage bio-inspired controller for a future six-legged robot. It combines a software representation of Drosophila neural connectivity with a real-time motor core and low-latency edge inference. The aim is to study how sensory perception, memory, and motor decisions can cooperate on affordable heterogeneous hardware.

Architecture: a GTI2801 accelerator processes visual frames; a Coral Edge TPU runs a quantized temporal sensor model; and a 500 Hz motor BioCore produces gait-control signals. A separate software MaleCNS graph represents 167,565 neurons and 25,623,478 synaptic edges. These components are being validated independently and in shadow-mode experiments.

Measured component-level results include GTI2801 vision inference around 6.26 ms and Coral Edge TPU temporal inference around 0.31 ms in separate tests. Shadow-mode evaluation observed motor-state differences in approximately 4.54% of 400 ticks. These are research measurements, not proof of a completed end-to-end robot.

Status: working software prototypes and accelerator experiments exist. The dedicated controller PCB is at the requirements/pin-map stage; it has not been fabricated or electrically validated. A completed physical hexapod is not claimed. The next milestones are reproducible integration tests, PCB design review, and physical robot validation.

The project is entered in Edge AI & IoT because it explores robotics, embedded inference, and practical hardware/software co-design. Third-party silicon, datasets, SDKs, and models remain subject to their own licenses; this description does not grant rights to those materials.

Related Hugging Face resources: https://huggingface.co/AvaSiG/ARGOS-MUKHTAR (model and project documentation); https://huggingface.co/spaces/AvaSiG/ARGOS-MUKHTAR-Lab (interactive lab); https://huggingface.co/spaces/AvaSiG/ARGOS-MUKHTAR-ZeroGPU (experimental demo). These resources document the research prototype; they do not represent a fabricated PCB.