Are autonomous vehicles better off without signals?

The arrival of autonomous vehicles has prompted many to imagine a world without annoying traffic signals.  If cars are smart enough to drive themselves, one is inclined to reason, why cannot they simply chart a crash-free path through at-grade intersections all by themselves?  In this paper, we ask: is eliminating signals from  intersection desirable, even if it is possible?  The abstract below provides a short answer (hint: Yes and No).   If you are interested in reading the full paper, a preprint may be found here. The paper recently appeared in Transportation Research Part B.


Abstract: We model and analyze a futuristic intersection that serves only connected, autonomous and centrally managed vehicles. Under consideration are three control strategies that aim to minimize the total system delay by choosing an optimal trajectory for each vehicle. The first two abandon the concept of signal timing all together whereas the third strategy keeps it. The difference between the two signal-free strategies has to do with a fail-safe buffer requirement introduced to provide redundancy. Each control strategy leads to a unique version of a trajectory-based autonomous intersection management (T-AIM) problem, which is formulated as a mixed integer linear program and solved using a variety of techniques. We found the signal-free strategy holds an overwhelming advantage over the signal-based strategy in terms of efficiency. However, its success is fragile and dependent on the faith in the safety and reliability of the system. When the fail-safe buffer is introduced, the efficiency of the signal-free strategy degrades to a level comparable to that of a properly designed signal-based strategy. Surprisingly, the signal-free strategy with redundancy tends to arrange vehicles in groups that take turns to cross the intersection together. This “signal-like behavior” manifests itself whenever congestion rises to certain threshold. In addition, solving the T-AIM problem based on signal timing enjoys significant computational benefits, because it eliminates cross conflicts. Thus, the basic logic of signal timing—if not the physical equipment—may survive even after humans are no longer allowed to drive.

Social Limits on Growth

Fred Hirsch was no big-time economist.  His brief academic career was cut short by ALS, the same disease that killed Tony Judt. Yet, his Social Limits on Growth is a masterpiece, probably not a book you could read for fun lying on the coach, but definitely worth the time and effort.

According to Hirsch, capitalism is doomed to trap everyone in counterproductive competition for positional goods such as Ivy League education and elite jobs.  This phenomenon may be best described as “involution” (内卷), to borrow a popular Chinese Internet Meme.   Economic growth cannot solve the involution trap. On the contrary, growth is bound to fortify it, by fulfilling the ever-increasing demand for material goods. Nor could redistribution overcome the scarcity of position goods. As Hirsch noted, “there is no such thing as leveling up” when reward is set by the position on the slope, because “the slope itself prevents a leveling”.    Therefore, the image of “a rising-tide-lifts-all-boats” is an illusion because the tide cannot keep rising and not all boats could stay above the water at the same time.

Hirsch seems rather pessimistic about finding any operational solution to the social limits on growth. In the end, he wonders whether the belief in incremental progress itself is but a pipe-dream, “a nonfiction version of the happy-ending”, or “a faith that is as utopian as the Utopianism it seeks to replace”.

Fall and rise of taxi travel during COVID

This is our second COVID19 related study, completed in 2020 and published in Transportation Research Part A in 2021. You may read  the other one here, which is about optimally adapting transit design and operations in a pandemic.

We examined taxi trajectory data collected in four weeks that cover the onset of COVID19, the shutdown, and phased reopening in the city of Shenzhen. Our analysis revealed how the pandemic and the travel restriction policies affected both the supply and the demand of the taxi market in the city.   One of the more interesting findings is that the city’s stimulus policy, designed to boost taxi supply and help taxi drivers, might have led to oversupply, by inducing taxi drivers to spend more time on the road than what the prevailing market condition would justify.  We uncovered direct evidence from data to support this finding through a clustering analysis.

A preprint is available here.


Abstract: This paper traces the plunge and rebound of the taxi market in Shenzhen, China through the COVID-19 lockdown. A four-week taxi GPS trajectory data set is collected in the first quarter of 2020, which covers the period of lockdown and phased reopening in the city. We conduct a spatiotemporal analysis of taxi demand using the data, and then select taxis that continued to operate through the analysis period to examine whether and how they adjusted operational strategies. We find, among other things: (i) the taxi demand in Shenzhen shrank more than 85% in the lockdown phase and barely recovered from that bottom even after the city began to reopen; (ii) the recovery of taxi travel fell far behind that of the overall vehicle travel in the city; (iii) most taxis significantly cut back work hours in response to the lockdown, and many adjusted work schedule to focus on serving peak-time demand after it was lifted; (iv) taxi drivers demonstrate distinct behavioral adaptations to the pandemic that can be identified by a clustering analysis; and (v) while the level of taxi service dropped precipitately at the beginning, it quickly rebounded to exceed the pre-pandemic level, thanks to the government’s incentive policy. These empirical findings suggest (i) incentives aiming at boosting supply should more precisely target where the boost is most needed; (ii) the taxi market conditions should be closely monitored to support and adjust policies; and (iii) when the demand is severely depressed by lockdown orders or when the market is oversupplied, taxi drivers should be encouraged and aided to use more centralized dispatching modes.

How can the taxi industry survive the tide of ridesourcing?

This  paper makes two empirical findings and one prediction. First, it reveals the intensity and scope of the impact of ridesourcing on the conventional taxi industry. Second, it uncovers evidence that taxis may be competitive in densely populated areas.  The second finding leads to a follow-up study you can read here.

I predict that the taxi industry is here to stay in the foreseeable future.    Here is what I wrote in the conclusion:

“Beyond e-hailing, economy of scale and aggressive pricing, ridesourcing does not seem to have other means at present to drive its expansion in the market. E-hailing is no longer the secret weapon that once glorifies the cause of TNCs – it can be easily picked up by a taxi dispatcher that owns and operates its own fleet. Aggressive pricing, on the other hand, has proven at best a double-edged sword, as Uber’s recent bitter defeat in China has vividly demonstrated. The scale of TNCs, which gives outside visitors a brand to stick to, is indeed an important competitive advantage. Even this lead is not that difficult to catch up, however, if a mobile platform, presumably operated by a third party, can unify taxi dispatchers around the world. Such a platform can easily work within cities’ existing regulatory structure, rather than against it, because it utilizes a dedicated and existing fleet. It can also improve the experience of street-hailing, a decisive advantage it holds against ridesourcing, by offering customers the amenities considered only available to e-hailing users, such as paying the fare on-line and rating drivers, all in real-time. An obvious solution may be allowing customers, as they board the taxi hailed off street, to open up an electronic transaction session similar to those seen on e-haling platforms, by e.g. scanning a QR code attached to the taxis or the driver’s smart phone.”… therefore, “The revolution of ridesourcing is unlikely to eliminate the necessity of a dedicated service fleet, and for years to come we will continue to live in a world with both ridesourcing and (upgraded) taxis.”.

The Journal of Transportation Research Part C selected this paper to receive the Best Paper Award in 2018. You may download a preprint here.


Abstract:  This paper aims to examine the impact of ridesourcing on the taxi industry and explore where, when and how taxis can compete more effectively. To this end a large taxi GPS trajectory data set collected in Shenzhen, China is mined and more than 2,700 taxis (or about 18% of all registered in the city) are tracked in a period of three years, from January 2013 to November 2015, when both e-hailing and ridesourcing were rapidly spreading in the city. The long sequence of GPS data points is first broken into separate “trips”, each corresponding to a unique passenger state, an origin/destination zone, and a starting/ending time. By examining the trip statistics, we found that: (1) the taxi industry in Shenzhen has experienced a significant loss in its ridership that can be indisputably credited to the competition from ridesourcing. Yet, the evidence is also strong that the shock was relatively short and that the loss of the taxi industry had begun to stabilize since the second half of 2015; (2) taxis are found to compete more effectively with ridesourcing in peak period (6-10 AM, 5-8 PM) and in areas with high population density. (3) e-hailing helps lift the capacity utilization rate of taxis. Yet, the gains are generally modest except for the off-peak period, and excessive competition can lead to severely under-utilized capacities; and (4) ridesourcing worsens congestion for taxis in the city, but the impact was relatively mild. We conclude that a dedicated service fleet with exclusive street-hailing access will continue to co-exist with ridesourcing and that regulations are needed to ensure this market operate properly.

Winners take all

Winners take all is about the dream of “doing well by doing good”, the idea that there is always a win-win solution to every social problem, and the belief that elites equipped with technology and market tools should be entrusted to lead social changes, preferably independent of the democratic processes, or politics. The author argues these ideas are largely an illusion, if not a deception, because the overlap between individual and collective interest is limited.  There is nothing wrong about wanting to do well by doing good; but we are kidding ourselves if we believe these do-gooders are saviors of our collective future.  The book is very harsh on the intellectuals who promote these ideas.  These so-called “thought leaders” are described as the cheerleaders employed by the “idea industry” to project positive energy, presumably at the expense of our collective good.

Overall, a good and easy read, and the main argument is fair and well-reasoned, if not entirely neutral.

To pool or not to pool

To Pool or Not to Pool: Equilibrium,  Pricing and Regulation

This paper was the first published based on  Kenan’s PhD research. It introduces ride-pooling into the equilibrium analysis of the ride-hail market and analyzes the effect of pricing strategies and various regulations on pooling.

After the first draft is completed in the Spring of 2019,  it took almost two years  to move the paper through various stages of the review process, first at Management Science, then at Transportation Research Part B (three rounds). While the long waiting was no doubt frustrating, the quality of the paper might have benefited from intensive scrutiny and repeated revisions.  For a preprint, please check here; the link to the final version is here.


Abstract: We study a monopoly transportation network company (TNC) in an aggregate market that offers on-demand solo and pooling e-hail services, while competing with transit for passengers. The market equilibrium is established based on a spatial driver-passenger matching model that characterizes the passenger wait time for both solo and pooling rides. We prove, under mild conditions, this system always has an equilibrium solution. Built on the market equilibrium, three variants of pricing problems are analyzed and compared, namely, (i) profit maximization, (ii) profit maximization subject to regulatory constraints, and (iii) social welfare maximization subject to a revenue-neutral constraint. A comprehensive case study is constructed using TNC data collected in the city of Chicago. We found pooling is desirable when demand is high, but supply is scarce. However, its benefit diminishes quickly as the average en-route detour time (i.e., the difference between the average duration of solo and pooling trips) increases. Without regulations, a mixed strategy—providing both solo and pooling rides—not only achieves the highest profit and trip production in most scenarios, but also gains higher social welfare. The minimum wage policy can improve social welfare in the short term. However, in the long run, the TNC could react by limiting the size of the driver pool, and consequently, render the policy counterproductive, even pushing social welfare below the unregulated level. Moreover, by maintaining the supply and demand of ride-hail at an artificially high level, the minimum wage policy tends to exacerbate traffic congestion by depressing the use of collective modes (transit and pooling). A congestion tax policy that penalizes solo rides promotes pooling, but consistently harms social welfare. However, it promises to increase both social welfare and pooling ratio, when jointly implemented with the minimum wage policy.

Travel demand management practice in China

This article stemmed from a research report commissioned by World Bank back in 2019.   I’ve never tried to publish it in a journal, though my friend Daizong Liu had helped translate it into Chinese and published it online through his very successful Wechat public platform (一览众山小).  The article reviews the travel demand management practice in China and attempts to draw some useful lessons from it.  You may read the abstract below and download the Egnlish version at ChinaTDM.


Lessons Learned from China’s Travel demand management practice

China’s car ownership has been expanding at a staggering pace in the past two decades. The rapid motorization brought unprecedented level of traffic to its densely populated cities
unprepared to accommodate it, causing severe congestion and air pollution problems. Chinese cities have responded to these challenges with sweeping travel demand management (TDM) measures. The practice of TDM in China is unique not only because it is large in scale and broad in scope, but also because it occurs against the backdrop of a fast and historical transition of the most populous country on earth. The objective of this note is to review and document this practice, discuss its outcomes and lessons, and examine what the rest of the world could learn from it.

Antifragile

This short review was originally written in April of 2021.


Overall,  Antifragile is a disappointment.  This is not to say it offers no interesting and useful ideas. It does.  What strikes me the most are the oversized impact of tail events (black swans) and their utter unpredictability, our ruinous obsession with optimization and intervention, and the agent problem ubiquitous in modern societies.  However, Taleb could have explained these ideas in 40 pages. Instead, he wrote 400, filling many of them with impulsive bragging, as well as his signature rant against the entire intellectual establishment.  In the end, I felt these self-inflicted distractions severely undermine the narrative and the logic flow.    I was looking forward to reading Black Swan, but after this experience, I wonder whether it would be worth my time.

A special note for my fellow academics who might be interested in Taleb’s work: he absolutely hates professors and minces no words berating them, so if you take the challenge, buckle up for a bumpy ride.

Hyperbush

Hyperbush Algorithm for Strategy-based Equilibrium Traffic Assignment Problems

This recent paper extends the concept of bush to hyperbush and uses it devise a new class of algorithms for strategy-based traffic equilibrium assignment problem, of which frequency-based transit assignment is an archetype.  One of the longest papers that I’ve ever written (nearly 50 pages), the paper just appeared online in Transportation Science.    Here is a preprint  Hyperbush.


Strategy-based equilibrium traffic assignment (SETA) problems define travel choice broadly as a strategy rather than a simple path. Travelers navigating through a network based on a strategy end up following a hyperpath. SETA is well suited to represent a rich set of travel choices that take place en-route at nodes, such as transit passengers’ transfer decision, truckers’ bidding decision and taxi drivers’ re-position decision. This paper recognizes and highlights the commonalities among classical and emerging SETA problems and proposes to unify them within a same modeling framework, built on the concept of hypergragh. A generic hyperbush algorithm (HBA) is developed by decomposing a hypergraph into destination-based hyperbushes. By constructing hyperbushes and limiting traffic assignment to them, HBA promises to obtain more precise solutions to larger instances of SETA problems at a lower computational cost, both in terms of CPU time and memory consumption. To demonstrate its generality and efficiency, we tailor HBA to solve two SETA problems. The results confirm HBA consistently outperforms the benchmark algorithms in the literature, including two state-of-the-art {hyperpath-based} algorithms. To obtain high-quality equilibrium solutions for SETA instances of practical size, HBA runs up to five times faster than the best competitor with a fraction of its memory consumption.