1. How I Invented RLADP, The Most Advanced Intelligent Control, Now Used In Iron Dome

 

1.2 From Harvard to LSE to DOE to NSF

 

In 2022, I received the Frank Rosenblatt award, its highest technical field award for computational intelligence. They cited two inventions – backpropagation  and RLADP.

 

I will never forget the day when I was a passenger in a Volkswagen going to Wellesley, when I realized that I could translate BOTH ideas from Freud’s theory of psychodynamics into workable mathematics. The first FORTRAN implementation of backpropagation came in 1967-68, when I was at LSE. Back at Harvard, my PhD thesis proved the “chain rule for ordered derivatives,” which I generalized to recurrent networks and optimization when I was at DOE 1979-1989.

 

My first act as a research program director in Engineering  at NSF (1988-2015) was to organize the workshop in New Hampshire which led to the book Neural Networks For Control, MIT Press 1990, edited by Miller, Sutton and Werbos. My two chapters spelled out a roadmap for that new field of research, including a “ladder” of designs for RLADP.

 

IEEE conferences in 1987 and 1988 gave great publicity to the new field of neural networks for engineering and for IT in general. Because of my contributions in that field, NSF hired me to define and manage their first program in “Neuroengineering” (and also to update and run a new program in Emerging Technologies Initiation,  very crossdisciplinary open area).

 

1.3. From 1988 to 1995: Research and Discussions Enabling Iron Dome

 

Among the key players whom I funded from NSF were Professor S. N. Balakrishnan (Bala), from U. of Missourri Rolla (later called MST), and David Hyland, from Harris Semiconductor to haed of aerospace at U. Michigan to Texas A&M.

 

I was part of the group in Engineering who created and ran a special research initiative in Intelligent Control. I “met” Bala when reading his proposal to that initiative. He proposed to develop and apply Potryagin’s design for optimization problems. I do not know what the reviewers decided, but I contacted Bala directly, and informed him to the very close connection between Pontryagin and my own new RLADP design, Dual Heuristic Programming (https://werbos.com/HIC_Chapter10.pdf). Bala submitted proposals to me in RLADP which I funded. He is the one who made Iron Dome work.

 

I was also grateful to meet David Hyland when he was using neural networks for work at Harris which I learned about from NASA. I funded him for more advanced neutral network work, which broke records in high quality high resolution imaging. By 1995, the Air Force was fully aware of that work, and chose Hyland to organize a sweeping, global conference at Albuquerque Air Force Base, with presentations by all the leaders they could find with tools and methods for imaging applications relevant to them. Part of the grant was to get them hard quantitative evidence about which methods work best. We all found it convincing and honest that Hyland himself had the result results; that is why they chose him, and why we funded him several times under strict peer review.

 

I will never forget that incredible week in 1995, held in a large windowless building which reminded me of UFO stories. That week was not long after big stories in the newspapers about the great success of the Patriot missile in the Middle East. I was grateful to have a nice quiet meeting with the high officer responsible for that area in that building. I said to him: “I read that Patriots have been an important life-saver, but they only have about 10-20 % success in hit to kill interception defense. We at NSF have developed/funded new intelligent control designs which can do much better. You and I should be working together somehow.”

I will never forget his angry body language, shouting: “WRONG on two counts. First, the Patriot missiles do not have 10-20% success. It is more like 1%. You can understand why we don’t want people to see that in the newspapers. Second, we cannot work on improving that because it is officially a solved problem. That means that Raythoen gets lots of money to solve it, and it would be an insult and a violation  of national security to rely on anyone else.”

 

1.4. Bala Wins and Keeps Moving – from 1995 to me at Senate Bldg 2009

 

But as it happened, I knew a high-ranking officer in the Air Force whom I had met many times at neural network conferences. He was promoted to BMDO, and we discussed the issue. Somehow he connected me to a major “shoot-out” project managed by Cottrell of Nichols Research, using AF test data and problems to compare competing designs for hit to kill missile defense. Bala won hands down.

 

Naturally, Bala was funded to do more, and contributed chapters to books on RLADP I funded and helped organize. We had many friendly conversations at many places. I even had some contact with his family, like his brother who created a major new university in New Delhi. I will never forget my last conversation with him in the Westin in Arlington, close to the Bezos headquarters east which we often walk by even now.

 

In 2009, I worked for a year as the science officer for Senator Specter, on the Defense Appropriations Subcommittee. I will never forget my meetings at the Marshall Institute, located in the Republican Club just south of the Capitol. At one of those meetings, covering missile defense, the Lockheed big representative walked over to talk with me, with an MIT professor in tow. The MIT guy certainly knew of Bala; he was VERY grateful for what Bala gave them, being studied and used somehow in their group at MIT. “We still don’t fully understand how it works, but the basic details are now working fine and creating a revolution.” THAT WAS IRON DOME – phase one.

 

At the Westin, Bala later told me that Lockheed and MKIT never understood more than stage one. That was all they needed to have a really great cash cow and educate many students. However, HE had provided them with a stage one AND a stage two design, which would knock out 90% of the missiles which would have made it through. By this time, he had a very serious group at his engineering university, and had lots of tests and proofs. Lockheed never followed up on it. What’s more, he had health problems, and that meeting at the Westin was the last I saw of him.

 

2. Beyond And Above Iron Dome

However, this was not the peak of this GNC technology!!!

Bala’s group was very strong. His group manager knew the operational aspects even more than Bala himself, and led the work of Iron Dome stage two. He is now in India, and I know where to find him. We only exchanged a few emails, but I believe he would be very happy to honor Bala’s cause – if we are honorable enough ourselves.

But this DHP technology based on classical physics is not the top of that pyramid. If you search on the patent for tQuA, and locate the papers already posted by USPTO, you will see: (1) claims about a much more advanced RLADP, which could be implemented in just a few months or even less if implemented on any of the quantum simulators we know about, the best of which is Pasqal in France; (2) claims in a paper by Hyland and Werbos, offering orders of magnitude better performance (e.g in Pasqal) in “seeing the sky” (or the earth), a crucial element of missile defense and warning in general, whether for military or civilian use. I often speak up in support of the Un Security Council session in 2023, which concluded that an open transparent international system for global warning is humanity’s number one need to prevent extinction by warfare.