SDK Tool Focuses on Symbiotic AI

As artificial intelligence (AI)-driven tools are integrated into system design processes, human-machine collaborative workflows have to navigate possible biases of human designers while promoting human insights, according to Charles River Analytics. An AI-driven alternative approach to the design process that addresses these needs is the impetus behind the Joint, Adaptive, Robust Visualization and Interaction System (JARVIS) tool from Charles River Analytics. JARVIS supports an iterative process of design and evaluation to help move cyber and physical subsystems through development cycles.

The 4-year contract was funded by the Defense Advanced Research Projects Agency (DARPA) under the Symbiotic Design for Cyber Physical Systems (SDCPS) program.

JARVIS banks on a symbiosis between humans and AI to serve two functions. First, it helps system designers explore the space of the possible to find new designs. Using the tool’s human-machine interface (HMI), designers can provide feedback to the AI to guide its exploration of the design space. Second, JARVIS helps designers refine specific components in new or existing designs to fine-tune each design’s ability to meet requirements. The tool provides a way for designers to understand the results of AI.

The advantage of using AI‑human symbiosis as the basis for the tool is that it harnesses both. AI can churn through millions of design candidates and conduct early performance assessments to move designers toward new and potentially promising design directions, according to Charles River Analytics. On the other hand, humans understand the limitations and practicality of the designs and use intuition to guide the design process.

JARVIS lets designers explore designs, then go back to previous designs and try different choices, and merge ideas from two streams of exploration. JARVIS enables optimization for any relevant parameter, whether material cost, speed and maneuverability of a vehicle, weight, or power consumption. The tool is agnostic to the specific domain of the design, so it is equally suited to exploring designs for air vehicles, underwater vehicles, or other systems.

Early use cases of JARVIS for DARPA explored the “correct-by-construction” design of cyber-physical systems like aerial and underwater vehicles. 

David Koelle, principal software engineer at Charles River Analytics and product manager on JARVIS. sees commercial potential for JARVIS, given its position as a tool focused on the designer’s experience and communicates useful feedback to AI tools.

The JARVIS software development kit (SDK) can work with in-house data and/or AI and machine learning tools. JARVIS supports existing design workflows and helps designers optimize simple designs and complex designs that involve multivariate parameters.

“In this new era of AI-driven design, JARVIS makes sure that design engineers have the right tools to deliver creative, efficient, and cost-effective systems,” Koelle says. With JARVIS, AI‑human symbiosis can help achieve correct-by-construction designs that also value innovation.

Sources: Press materials received from the company and additional information gleaned from the company’s website.