SiMa.ai opens Model Browser access to physical AI measurements
US chip developer SiMa.ai has opened a Model Browser presenting measurements for AI used in robots and drones. It exposes operating conditions for perception, depth and reasoning workloads on its Modalix chips. The company-supplied results can be examined with memory, power and execution conditions in view when considering models for these systems.
Artificial Intelligence··Night
Model Browser exposes operating measurements
US chip developer SiMa.ai has opened Model Browser access to measurements for physical AI, software used to perceive surroundings and act in robots or drones. Its November 17 disclosure concerns access to these results. The Palette software development kit was distributed earlier. The new browser presents measurements for perception, depth and reasoning workloads using the company’s Modalix chips.[1], [2]
Multiple models share the same chip
Palette can run multiple models concurrently on a chip, including workloads separated into Docker containers, isolated software environments. Language and vision-language models share constrained memory and power budgets. The software changes concern work performed on existing Modalix hardware. Measurements can be examined with operating conditions in view, including memory, power and execution settings. The published results are measurements supplied by SiMa.ai.[1]
Sensor and simulation work provides context
SiMa.ai also describes work with ecosystem partners on combining sensor information, estimating depth and using digital twins in closed-loop simulations. Sensor fusion brings different sensor inputs together; depth processing represents distance in the scene. Digital twins provide simulated environments for the system’s actions and feedback. This work provides technical context for the perception and reasoning workloads represented in Model Browser.[1]