How Rainmaker, Prometheus, FIRESTORM, and SHOT AI algorithms enable the Kill WebBlog
December 23, 2021
WARFARE EVOLUTION BLOG. In our previous adventure, I introduced the pyramid model to identify each service’s Kill Web program and their experimental activities. Sitting at the peak is the Pentagon’s JADC2 program (Joint All Domain Command and Control) and the GIDE-XX exercises. On one side is the Army's IBCS program (Integrated Battle Management System) and Project Convergence exercises. On the next side is the Navy’s CEC program (Cooperative Engagement Capability) and their highly classified Project Overmatch exercises. On the next side is the Air Force's ABMS program (Advanced Battle Management System) and Onramp exercises. Since the Space Force stills falls under the Air Force, we’ll call their program SF-ABMS. They make up the final side of the pyramid. Space Force has been conducting a number of experiments under different names so we’ll decode those in the future.
For this article, we’ll examine the recent U.S. Army Kill Web exercises and implications. Project Convergence-21 (PC21) was conducted over 6 weeks here in the Arizona desert in October and November. PC20 (held in August 2020) concentrated on connecting several ISR (intelligence, surveillance, and reconnaissance) sensors and weapons together on a network, to reduce the time between detecting a target and destroying it (from 20 minutes down to 20 seconds). That exercise was primarily an Army platform affair, testing 34 new technologies and concepts. PC21, however, concentrated on the network itself, connecting multiple ISR and weapons platforms from the Air Force, Navy, Marine Corps, and Space Force into the Army’s Kill Web mesh network to produce a detailed realtime COP (common operating picture) using 110 new technologies and concepts. As you would suspect, there were some problems.
What the Army accomplished at PC21 is multidimensional and complicated, so the best way to view the results is through the four AI (artificial intelligence) algorithms employed in the exercise: Rainmaker, Prometheus, FIRESTORM, and SHOT. Otherwise, I’ll start rambling and this article will turn into a convoluted mess. The scarcity of information about these AI modules forced me to make some logical deductions here and there, so this is how I think they worked at PC21.
The first thing the Army needed to do was set-up the network and connect all the different sensors and weapons systems, from all the different services involved, into a secure mesh network and start collecting ISR data. That included radar, IR (infrared), electro-optical (imaging), and RF-intercept sensors. However, the data coming from all those sensors operate on different radio frequencies, so the network must discover and establish those frequency-appropriate links.
To read more Warfare Evolution Blogs by Ray Alderman, click here.
Additionally, the data coming from those sensors use different message formats and data structures, since many of them are legacy systems. Somewhere on the network, the data must be translated into a common message format and data structure. That’s what Rainmaker did at PC21. It connected 15 sensors, 19 weapons, the combat cloud servers at Joint Base Lewis-McChord, military and commercial satellites, units at White Sands Missile Range, soldiers on the ground at Yuma Proving Grounds, crews at Vandenburg SFB (Space Force Base), Army air and missile defense systems, some combat ground vehicles, some ISR drones, some fighter planes, some helicopters, some artillery, and a few other military bases.
Once the network was up and running, another Army group played the part of the enemy and interfered with the RF frequencies, satellite communications links, and satellite PNT (Position, Navigation, and Timing) signals in four ways: denied access, degraded access, intermittent access, and limited bandwidth access. They did this by using their electronic warfare (EW) systems. That forced Rainmaker to implement several new jam-resistant waveforms for the RF and satellite links, and find open communications channels using the PACE model (primary, alternate, contingency, and emergency).
The network reconfigured and found new paths for the sensor data. Imaging data from a small UAV (unmanned aerial vehicle) was sent to another larger UAV, that sent it to a Blackhawk helicopter or fighter plane, that sent it to a LEO (low Earth orbit) satellite, that sent it to the combat cloud servers at Lewis-McChord. So, it looks like there is a "guaranteed forward progress" communications algorithm running inside Rainmaker. Otherwise, the data from a sensor could ping-pong between different platforms and never get to the combat cloud servers. Obviously, the EW attack reduced bandwidth and induced latencies on the mesh network as the data was relayed on multiple paths through multiple platforms. Rainmaker figured out the new paths and frequencies autonomously and dynamically in seconds.
Now that the ISR data from the different sensors is coming in at Lewis-McChord’s combat cloud servers, and has been properly structured and formatted, it’s time to fuse all that data together and analyze it. So, Rainmaker pushed the ISR data over to the Prometheus algorithm, whose job is to find threats and targets. Once targets are found, Prometheus pushes those target coordinates over to the FIRESTORM algorithm (Fires Synchronization To Optimize Responses in Multi-domain operations).
FIRESTORM also knows the positions and capabilities of the weapons connected to the network, so it begins to match the targets to the appropriate weapons using the decision models we have discussed previously: emergent coordination, Greedy Shooter, hierarchical coordination, central coordination, and consensus coordination. The match is based on proximity, the explosive power of the warhead needed to destroy the target, weather, geography, and other factors. At PC20 last year, a primitive version of FIRESTORM found six firing solutions for a target. At PC21, an enhanced version of the algorithm found 21 sensor-to-shooter firing solutions, mainly because more sensors and weapons were connected to the network. FIRESTORM did all this autonomously and dynamically in seconds.
FIRESTORM then pushed the firing solutions to the battlefield commander. Once the commander makes a sensor-target-weapon selection, that decision is sent to the SHOT algorithm (Synchronized High Optempo Targeting). At that point, SHOT disconnected all the unselected weapons from the firing sequence and fired the selected weapon. Sensors and weapons are matched to targets and selected weapons are fired in seconds, so the available weapons pool on the network needs to stay current since FIRESTORM could have hundreds of targets that need weapon assignments. SHOT’s job is to manage the firing sequence (sensor to shooter), and possibly the weapons pool to some degree.
Once the selected weapon was fired at the target, SHOT and FIRESTORM backed off and Prometheus took over, doing BDA (Battle Damage Assessment) with new data from Rainmaker. If the target was not destroyed, then we go through the Rainmaker-Prometheus-FIRESTORM-SHOT sequence again and fire another weapon. But, don’t worry. That only takes a few seconds.
More information about PC21 is slowly trickling-out of the Army, but much of the results are classified. The Army did fire their PrSM (Precision Strike Missile) in the desert, from a mobile ground launcher. The missile has a range of more than 300 miles. Maybe it just travelled a short trajectory of 40 or 50 miles and hit a target at the Yuma Proving Grounds. As for other weapons fired at targets, there’s not much detail in the reports I have read.
The Army launched some small ISR drones out of missile tubes, one from a helicopter and one from a ground combat vehicle. They were tasked with forward reconnaissance missions and they worked. Four small quadcopters were matched with four autonomous ground combat vehicles and told to explore a certain area in the desert and report back on the network. So, they did it. Other autonomous ground vehicles were tasked with delivering supplies to a tank in the field, and that was accomplished. The COP was shared over the network with higher command levels. But, PC21 was primarily a tactical network experiment, connecting from the brigade level down through battalion, company, platoon, squad, and fire team. It was not designed to connect from the brigade level up through division, corps, field army, army group, and Pentagon. And an F-35 fighter plane sent data about enemy targets directly to troops on the ground in Yuma, without going up to Lewis-McChord and back on a satellite link.
Back in 2017, the Army activated their first MDTF (Multi-Domain Task Force), installing the combat cloud servers (running Rainmaker, Prometheus, FIRESTORM, and SHOT) along with the satellite links at Lewis-McChord in Washington State. This was mostly for Project Convergence exercises and testing. In September 2021, the Army announced the activation of the second MDTF in Wiesbaden, Germany. They are planning for a total of five MDTF locations: two in the Pacific, one in Europe, one in the Arctic, and another for global response. Lewis-McChord could handle the North Pacific, and the second Pacific MDTF could be in Hawaii, to handle the Southern Pacific. The Wiesbaden MDTF will handle Europe. Alaska would be a nice location for the Arctic MDTF. The Global MDTF could be an air-mobile unit, that could pop-up Kill Webs in 24 hours anywhere in the world in a crisis.
Now, compare the AOR (area of responsibility) of these MDTFs to the six U.S. military regional combat commands. Two facing the Pacific (Lewis-McChord and Hawaii) could establish Kill Webs for U.S. INDOPACOM (Indian Ocean-Pacific Ocean Combat Command), in case China tries to invade Taiwan. INDOPACOM is located at Camp Smith, Hawaii, which supports my theory the one of the Pacific MDTFs could be located there.
One MDTF facing Europe and another facing the Arctic (Wiesbaden and Alaska) could establish Kill Webs for EUCOM (European Combat Command) against Russia invading Europe. EUCOM is located in Stuttgart, Germany, and the recently announced European MDTF is located in Wiesbaden, just down the road.
The Global MDTF could pop-up a Kill Web in the United Kingdom and support U.S. CENTCOM (Central Command) if Iran attacks Israel. And, they could deploy to Japan and set-up a localized Kill Web for INDOPACOM if North Korea invades South Korea. That makes sense to me, about where the remaining MDTFs could be located.
The Army is planning to deploy enhanced versions of Rainmaker, Prometheus, FIRESTORM and SHOT at the activated MDTFs starting in 2023. These AI algorithms will probably be running on commercial cloud servers under JWCC (Joint Warfighting Cloud Capability) now that the JEDI (Joint Enterprise Defense Infrastructure) contract is dead. The Army doesn’t need to maintain their own combat data centers when Google, Microsoft, Amazon, and others can do it more efficiently. Looks like we’ll see the Kill Web concept go operational soon.
There you have it. The Army learned that the network bogs down under certain conditions with just 34 primary nodes and is not scaleable. They identified hotspots, where certain sensors hog the bandwidth, and the network needs to be made more resilient to EW attack. That says that Rainmaker needs a lot of work, to handle hundreds or thousands of nodes. That can be done with more airborne communications nodes, and by doing more pre-processing of the data at the sensor (edge processing). The other AI modules seemed to work well, but they may need further refinement. What we saw at PC21 was the animation of the 5F Kill Web model: Find (identify), Fix (track), Fire, Finish, and Feedback.
Next time, we’ll take a look at one of the other services’ Kill Web experiments and their AI algorithms. Which one depends on how much reading I can do over the holidays. In the meantime, do some reading on Rainmaker, Prometheus, FIRESTORM, and SHOT. My analysis may not be perfect here, but what these AI algorithms do is amazing.