Satellites have long been expected to act as autonomous, self-sufficient devices. After all, they’re alone in space.
Yet, the operation of satellites is still largely dependent on crews at ground center, monitoring and often controlling the behavior of the device. Furthermore, satellites generally don’t process the data they’re transmitting or receiving.
This eliminates their ability to independently make real-time decisions with the communicated data.
Example of the various support necessary to maintain a satellite. Image used courtesy of a.i. solutions
Space technology company C3S is well aware of these shortcomings and is on a mission to correct them.
The Plan: Take Self-Driving Car Technology to Space
In an effort to create autonomous satellites capable of making real-time decisions, C3S has decided to team up with AImotive, a leader in autonomous vehicle technology.
AImotive is a multifaceted company in the autonomous vehicle space, providing a stack of AI-specific software, development and validation tools, and hardware. C3S will be looking to leverage AImotive’s hardware, namely their aiWARE automotive neural network accelerator chip. NN acceleration technology is going to be crucial for satellites when trying to perform inference quickly and at low power.
aiWARE evaluation kits. Image used courtesy of AImotive
Some of aiWARE’s fundamental features include:
- A configurable, low-latency, and high-efficiency architecture
- Optimization of on-chip SRAM and DDR
- Patented data management techniques
AImotive claims that aiWARE is designed to accommodate the objectives of a vehicle on the road, giving it a unique architecture as opposed to other NN accelerators on the market.
What’s to be Gained by the C3S-AImotive Partnership?
There are many NN accelerators on the market, so you might ask why these two companies specifically joined forces.
Of all of the fields that incorporate AI, autonomous vehicle technology most closely matches the goals and constraints of an autonomous satellite. AI in autonomous vehicles, like in a satellite, requires the ability to make real-time decisions given unpredictable inputs while simultaneously meeting stringent size and power constraints.
Other popular places for AI, like data centers, run neural networks on large and power-expensive hardware, often lacking the ability, or necessity, to make real-time decisions.
aiWARE portfolio. Image used courtesy of AImotive
Unlike data center applications, autonomous vehicles and satellites that use AI have no tolerance for any system that needs to periodically clean up buffers, do garbage collection, or otherwise stop for maintenance.
In a whitepaper from AImotive on the importance of offloading data processing (PDF), the company explains, “For automotive embedded inference, the memory strategies used must be robust under all conditions, and work continuously without fail.” To this end, AImotive has secured many patents focused primarily on data flow and management.
The companies’ shared goals make AImotive’s expertise suited for space applications.
Space-Specific Limitations to aiWARE
That’s not to say that the two technologies are perfectly cohesive, and there are definitely hurdles to be overcome. aiWARE is going to need C3S’ expertise to help convert its products to ones that will be able to withstand the demanding habitat of space.
Sources of radiation in space. Image used courtesy of NASA and Analog Devices
Considerations in space that earth-bound vehicles seldom encounter include heat fluctuations and cosmic radiation.
From Autonomous Vehicles to Autonomous Satellites
The partnership between these two companies provides yet another example of how fully autonomous satellites may be a realistic possibility in the future. Satellite applications that require immediate action like natural disaster detection, weather forecasting and alerting, and cargo tracking will likely see huge benefits from AI being brought into space.
Featured image used courtesy of AImotive
Automotive- and space-bound designs often work well together because of similar harsh environments. Have you ever been surprised by how well one of your designs worked with an application it wasn’t intended for? Share your experiences in the comments below.