Commentary: Optimizing a truck fleet utilizing synthetic intelligence – FreightWaves

The views expressed listed here are solely these of the creator and don’t essentially characterize the views of FreightWaves or its associates.  

Author’s Disclosure: I’m not an investor in Optimal Dynamics, both personally or via REFASHIOND Ventures. I’ve no different monetary relationship with Optimal Dynamics.

On July 7 I began a sequence on AI in Supply Chain (#AIinSupplyChain). The first article within the sequence profiled Optimal Dynamics, a startup that has launched a product to routinely optimize operations for giant trucking fleets.

The second article on this sequence profiled Warren Powell, co-founder of Optimal Dynamics and a soon-to-be professor emeritus of operations research at Princeton University, after a 39-year tenure there. He can be the Founder and Manager of Princeton University’s Computational Stochastic Optimization and Learning Labs (CASTLE Labs).

Dr. Warren Powell, co-founder of Optimal Dynamics.

Throughout his profession, Powell has been on the forefront of researching and growing fashions and algorithms for stochastic optimization with sensible functions in transportation and logistics. His analysis at CASTLE Labs has been supported by greater than $50 million in funding. He has authored two books and an edited quantity of articles, and 250+ papers on decision-making beneath uncertainty with functions to the issues encountered inside industrial provide chains.

Powell is nearing completion of a brand new e book “Reinforcement Learning and Stochastic Optimization: A Unified Framework for Sequential Decisions.” His educational family tree and lineage contains 60 Ph.D college students and postdocs, 10 Masters college students and 200+ undergraduate senior theses.

In this installment of the AI in Supply Chain sequence, he explains why fixing actual world provide chain issues is so onerous for many AI methods, and hints at among the breakthroughs he has made in the course of the a long time over which he has spent conducting utilized analysis on stochastic optimization – a flowery approach of claiming optimization beneath uncertainty.

Driving a truck is a transfer of knowledge and skills by humans (at least now).
Driving a truck is a switch of data and abilities by people (a minimum of now).
(Photo: Jim Allen/FreightWaves)

What are high-dimensional determination issues?

High-dimensional issues are issues for which the related information on which selections have to be made has tons of, probably hundreds of attributes, or dimensions – in plain English, the variety of variables is unmanageable. These attributes or dimensions are topic to randomness, and so they’re handled as portions possessing each a magnitude and a path – vector portions. Such information is analyzed utilizing algorithms designed particularly for this goal.

Modeling a truck driver, then modeling a truck fleet

To mannequin a truck driver, you want a 15-dimensional attribute vector (location, domicile, hours of service, hazmat flags, citizenship, tools traits, and many others). The spatial dimension rapidly introduces tens of hundreds of dimensions (consider metropolis pairs, for instance), which is then magnified when one has to think about the attributes of drivers, tools, the forms of merchandise being shipped, and different related information and knowledge.

Computational complexity will increase with high-dimensional information as a result of the variety of potential mixtures scales exponentially. This is the place Powell and the researchers he has been collaborating with at CASTLE Labs have made the breakthroughs that Optimal Dynamics is now bringing to market.

In the state of affairs with the truck driver above, Powell estimates that one will get 1020 completely different mixtures of attributes – consider a “1” with 20 zeros after it. Optimal Dynamics makes use of a way referred to as variable dimensional studying to get estimates of all of those values, and doesn’t want a ton of knowledge to do that.

As Powell places it, “This is what makes solving dynamic resource allocation problems so hard, high-dimensionality and uncertainty.” (See the July 7 and July 17 installments of #AIinSupplyChain for an evidence of dynamic useful resource allocation issues – the hyperlinks are on the finish of this text.)

Two truck drivers working together in a truck cab.
Two truck drivers working collectively.
(Photo: Jim Allen/FreightWaves)

During a dialog in October 2019, Paul Hofmann, who holds a Ph.D in nonlinear quantum dynamics and chaos idea from Technische Universität Darmstadt in Germany, and is the Chief Innovation Officer of the Alpega Group and INET Logistics, instructed me {that a} distinctive function of Optimal Dynamics’ platform,, is that it permits really dynamic optimization of logistics firms’ operations over completely different time horizons, from long-term via mid-term to real-time, all the time taking modifications and uncertainties under consideration.

Hofmann has a singular perspective on this matter as a result of he was the chief expertise officer of two AI and information science startups: Saffron Tech, a cognitive computing firm acquired by Intel; and SpaceTime Insight (now Nokia), an AI firm in power logistics.

He collaborated with Powell’s CASTLE Labs throughout his tenure as vp of R&D at SAP Labs in Palo Alto, California. He sponsored CASTLE Labs to discover functions of dynamic useful resource allocation within the power trade.

Cognitive methods promise to assist In decision-making, however the actuality is muddled

In the April 23, 2019 commentary, Logistics network optimization – why this time is different, I raised the subject of cognitive methods and the way such methods are beginning to be utilized in provide chain optimization.

According to the web site for the Cognitive Research Lab at Ulster University within the U.Okay., “Data analytics has evolved over the years from descriptive (what has happened) to diagnostic (why did it happen) to predictive (what could happen) to prescriptive (what action could be taken). The next big paradigm shift will be towards cognitive analytics. which will exploit the massive advances in high performance computing by combining advanced AI and machine learning techniques with data analytics approaches.”

It goes on to say that, “Cognitive analytics applies human-like intelligence to certain tasks, and brings together a number of intelligent technologies, including semantics, artificial intelligence algorithms, deep learning and machine learning. Applying such techniques, a cognitive application can get smarter and more effective over time by learning from its interactions with data and with humans.”

Nevertheless, because the Cofounder and Organizer of The Worldwide Supply Chain Federation and The New York Supply Chain Meetup, I’m listening to from members of the neighborhood that usually such superior, automated decision-making methods fail to ship on their promise, they usually encounter deadly setbacks when efforts are made to implement them in direction of fixing enterprise issues in sectors like manufacturing, healthcare, fast-moving shopper items, power, automotive, agriculture, transportation and logistics, and others.

Will AI be used by individual farmers in the near future? A Photo of a combine in a field.
Will AI be utilized by particular person farmers within the close to future?
(Photo: Jim Allen/FreightWaves)

This is corroborated within the enterprise press; on-line, in newspapers and magazines, and in books exploring the subject. For instance, within the June 11, 2020 version of its Technology Quarterly, the Economist revealed Businesses are finding AI hard to adopt.

While digital native firms like Facebook, Apple, Amazon, Netflix, Google, Microsoft, Baidu,, Alibaba, Tencent, Baidu, Tesla and different expertise upstarts use information analytics and AI to realize industrial energy, incumbent digital immigrant firms in legacy industries proceed to lag behind and lose market share as they wrestle and fail to unlock the potential and promise of knowledge analytics and AI.

Some of the explanations generally given for AI’s failure to proliferate inside industrial settings are: it’s troublesome to combine newer AI methods with present, mature methods and processes; firms in mature industries lack enough experience to implement such applied sciences or, if the experience is accessible, it’s too scarce and costly; executives don’t perceive these applied sciences; there’s a notion that the expertise is simply too immature; there’s inadequate information of the standard required to coach AI methods; and prior to now executives have been burned after shopping for into hype round AI. 

These and different hindrances to the adoption of AI in legacy industries like trucking are highlighted in Winning With AI, an October 2019 analysis report revealed by MIT Sloan Management Review, BCG Gamma, and BCG Henderson Institute.

According to that report, “Seven out of 10 companies surveyed report minimal or no impact from AI so far. Among the 90% of companies that have made at least some investment in AI, fewer than two of five report obtaining any business gains from AI in the past three years. This number improves to three of five when we include companies that have made significant investments in AI.”

The promise of AI is highlighted in Sizing The Prize, a 2017 report by PwC through which the authors estimate that AI will enhance international GDP by as much as 14%, or $15.7 trillion by 2030. However, in its 2020 AI Predictions, PwC finds that solely 4% of executives plan enterprise-wide deployments of AI. That is down from 20% the prior 12 months.

Optimal Dynamics is exclusive within the sense that, taken collectively, the members of its group have expertise confronting and fixing every of the issues that trigger AI implementations to fail as soon as such merchandise are introduced out of educational labs and into enterprise settings.

An IBM engineer works at a bank of servers...
Banks of servers being attended to…
(Photo: IBM)

How Optimal Dynamics’ is completely different from IBM Watson and Google DeepThoughts AlphaGo

While engaged on this text, this thought saved crossing my thoughts – in 1997 IBM’s Deep Blue beat Gary Kasparov, the chess grandmaster. In 2013, folks watched IBM Watson, the Jeopardy-winning super computer, beat Ken Jennings and Brad Rutter – two of one of the best Jeopardy gamers. Then in 2017 folks watched AlphaGo, the documentary about how a pc developed by DeepThoughts, now owned by Google, beat Lee Sedol, the 18-time World Go Champion in 2016.

To non-computer scientists the questions may come up – “Is Optimal Dynamics’ any different or better than IBM’s Watson or Google DeepMind’s AlphaGo? Can’t those systems do the same thing that can?”

According to Powell, “ has almost nothing in common with Watson or AlphaGo.”

“Watson is a machine learning system that must be trained using a large dataset. Watson uses a mixture of tools, but deep learning (with neural networks) is at the heart of the system. You need very large datasets to train neural networks. So, to get it to play chess, it had to be trained with a lot of history, but also millions of simulations of playing against itself. You can only do this type of training with a game (such as chess, or Go).” 

Deep studying is a subfield of machine studying through which algorithms are constructed utilizing layers of synthetic neural networks. Neural networks are mathematical buildings which mimic the traits and performance of neurons within the mind. Although neural networks have existed for the reason that 1950s, they’ve not too long ago regained recognition due to advances in algorithm design and computational energy.

A truck from the Hillshire Farm fleet is driven down the road.
A truck from the Hillshire Farm fleet is pushed down the street.
(Photo: Jim Allen/FreightWaves)

As the Jeopardy and Go examples reveal, deep studying algorithms now equal or surpass human efficiency on sure duties. AlphaGo combines deep studying and reinforcement studying. (See the July 17 instalment of #AIinSupplyChain for an evidence of reinforcement studying.)

He continued, “Otherwise, you need an external dataset, such as a set of patients with decisions made by doctors, or a set of radiology images that are labeled if they exhibit cancer.  But, these are all instances of what I call single-entity problems, comparable to optimizing a single truck. You could never use Watson to manage a fleet of trucks – it is simply the wrong tool for the problem.”

Elaborating additional, he stated “Our technology makes decisions – we need a ‘model’ that captures how the physical system evolves over time given the decisions we make (adding to inventory, dispatching trucks), and metrics that tell us how well we are doing.  We then use algorithms to choose among a family of functions called ‘policies’ which are methods for making decisions.” doesn’t want an enormous historic dataset of previous selections. That is a giant distinction. However, nonetheless must seek for one of the best coverage for making selections.

Elaborating additional on the place AI methods like IBM’s Watson and Google DeepThoughts’s AlphaGo would fail inside an industrial setting, Powell explains once more that, “A major characteristic of logistics is dimensionality. Tools such as Watson (or DeepMind’s tool for playing Go) are solving one-dimensional problems – what move to make. For Go, there are typically up to about 300 possible moves (depending on how full the board is). If we are managing 100 trucks, where each truck has a truck driver described by a 15-dimensional attribute vector, we can think of this as comparable to playing 100 games of Go all at the same time, and where we cannot make the same move for any two games. Our dispatch problem has a decision with 1,000 dimensions (if we only consider assigning a driver to up to 10 loads). The number of possible decisions is around 21000 = 10300 (think of a “1” with 300 zeros after it). And can deal with hundreds of vehicles.”

He concludes his rationalization of the excellence between Optimal Dynamics’ and IBM’s Watson and Google DeepThoughts’s AlphaGo by saying, “Handling these big problems is not as hard as it seems – we could handle these problems decades ago, as long as we ignored the impact of decisions now on the future.”

“Imagine planning a path to a destination without looking into the future. The problem with managing fleets and inventories is that you have to plan into the future, when you cannot predict the future (or even pretend that you can). It is this combination of uncertainty over time, and the dimensionality of the decisions, that makes these problems so hard,” he stated.

In quick, IBM’s Watson and Google DeepThoughts’s AlphaGo (and it’s descendant AlphaZero) can’t be utilized to those high-dimensional dynamic useful resource allocation issues, however can.

Watson, AlphaGo and AlphaZero excel in video games characterised by excellent info. In the actual world, within the industrial provide chains that matter, the place high-dimensional  and dynamic useful resource allocation issues are the rule, we by no means have excellent details about the longer term.


The Economist has declared AI dead as far as industrial applications are concerned. I want to reframe the query. In my opinion, AI isn’t lifeless for industrial provide chain functions. However, we have to rethink how we’re asking the questions, and subsequently to reassess how we’re going about fixing this class of issues.

Looking to the longer term, I imagine that synthetic intelligence and machine studying will contribute to the most important transformations in how selections are made in real-world industrial provide chains. However, there’s a have to get the analysis out of educational labs and into the actual world.

This #AIinSupplyChain sequence will proceed via the top of 2020, as I search to shine a lightweight on areas the place information, synthetic intelligence, and predictive analytics reside as much as their promise and potential for fixing high-dimensional determination issues by combining one of the best features of machine intelligence and human decision-making.

If you’re a group engaged on improvements that you just imagine have the potential to considerably refashion international, industrial provide chains we’d love to inform your story in FreightWaves. I’m simple to succeed in on LinkedIn and Twitter. Alternatively, you may attain out to any member of the editorial group at FreightWaves at

The reference archive dig deeper into the #AIinSupplyChain Series with FreightWaves

●     Commentary: Optimal Dynamics – the decision layer of logistics? (July 7)

●     Commentary: Combine optimization, machine learning and simulation to move freight (July 17)

●     Commentary: SmartHop brings AI to owner-operators and brokers (July 22)


Please enter your comment!
Please enter your name here