Ar­ti­fi­cial in­tel­li­gence

To­day we find our­selves in an­other trans­for­ma­tional era in hu­man his­tory. Much like the agri­cul­tural and in­dus­trial rev­o­lu­tions be­fore it, the dig­i­tal rev­o­lu­tion is re­defin­ing many as­pects of mod­ern life around the world. Ar­ti­fi­cial in­tel­li­gence plays an

Australian Transport News - - Contents - WORDS DHL A ND I BM

DHL and IBM re­lease a land­mark re­port, Ar­ti­fi­cial In­tel­li­gence in Lo­gis­tics

Global trans­port and de­liv­ery firm DHL and long-term com­put­ing gi­ant IBM have is­sued a land­mark col­lab­o­ra­tive re­port, Ar­ti­fi­cial In­tel­li­gence in Lo­gis­tics. It is a de­tailed, non-aca­demic anal­y­sis of the fu­ture of trans­port and lo­gis­tics and how it will get there.

In this edited ver­sion, an­a­lysts look at what ar­ti­fi­cial in­tel­li­gence (AI) is and how it will look to com­pany own­ers and man­agers. The ex­perts then ex­am­ine what it can do for cus­tomers and sup­pli­ers and where it is likely to go.

DHL se­nior vice pres­i­dent and global head of in­no­va­tion Matthias Heut­ger and IBM global in­dus­try leader for freight, lo­gis­tics and rail Keith Dierkx launched the re­port.

“To­day’s cur­rent tech­nol­ogy, busi­ness, and so­ci­etal con­di­tions favour a par­a­digm shift to proac­tive and pre­dic­tive lo­gis­tics op­er­a­tions more than any pre­vi­ous time in his­tory,” Heut­ger ex­plains. “As the tech­no­log­i­cal progress in the field of AI is pro­ceed­ing at great pace, we see it as our duty to ex­plore, to­gether with our cus­tomers and em­ploy­ees, how AI will shape the lo­gis­tics in­dus­try’s fu­ture.”

Dierkx notes that tech­nol­ogy is chang­ing the lo­gis­tics in­dus­try’s tra­di­tional value chains, and ecosys­tems are re­shap­ing en­ter­prises, in­dus­tries and economies.

“By lever­ag­ing AI into core pro­cesses, com­pa­nies can in­vest more in strate­gic

growth im­per­a­tives to mod­ernise or elim­i­nate legacy ap­pli­ca­tion sys­tems,” Dierkx says.

“This can make ex­ist­ing as­sets and in­fra­struc­ture more ef­fi­cient, while pro­vid­ing the work­force with time to en­hance their skills and ca­pa­bil­i­ties.”


In re­cent years, AI has come roar­ing out of re­search lab­o­ra­to­ries to be­come ubiq­ui­tous and am­bi­ent in our per­sonal lives, so much so that many con­sumers do not re­alise they use prod­ucts and ap­pli­ca­tions that con­tain AI on a daily ba­sis.

AI stands to greatly ben­e­fit all in­dus­tries, achiev­ing adop­tion leaps from con­sumer seg­ments to en­ter­prises and on­ward to the in­dus­trial sec­tor. Tech­no­log­i­cal progress in the fields of big data, al­go­rith­mic devel­op­ment, con­nec­tiv­ity, cloud com­put­ing and pro­cess­ing power have made the per­for­mance, ac­ces­si­bil­ity, and costs of AI more favourable than ever be­fore. Just as the re­la­tional data­base found its way into core busi­ness op­er­a­tions around the world – pro­vid­ing bet­ter ways to store, re­trieve, and or­gan­ise in­for­ma­tion – AI is now fol­low­ing a sim­i­lar path. It is be­com­ing an in­te­gral part of ev­ery fu­ture soft­ware sys­tem and soon we will no longer need to call it out as AI.

Al­ready to­day, AI is preva­lent in con­sumer-fac­ing ap­pli­ca­tions, cler­i­cal en­ter­prise func­tions, on­line and off­line re­tail, au­ton­o­mous mo­bil­ity, and in­tel­li­gent man­u­fac­tur­ing.

Lo­gis­tics is be­gin­ning its jour­ney to be­come an AI-driven in­dus­try, but the fu­ture is still rife with chal­lenges to over­come and op­por­tu­ni­ties to ex­ploit. With this in mind, ex­perts from IBM and DHL have jointly writ­ten this re­port to help you an­swer the fol­low­ing key ques­tions: • What is AI, and what does it mean for my

or­gan­i­sa­tion? • What best-prac­tice ex­am­ples from other

in­dus­tries can be ap­plied to lo­gis­tics? • How can AI be used in lo­gis­tics to rein­vent back of­fice, op­er­a­tional, and cus­tomer-fac­ing ac­tiv­i­ties? Look­ing ahead, we be­lieve AI has the po­ten­tial to sig­nif­i­cantly aug­ment cur­rent lo­gis­tics ac­tiv­i­ties from end to end. As in other in­dus­tries, AI will fun­da­men­tally ex­tend hu­man ef­fi­ciency in terms of reach, qual­ity, and speed by elim­i­nat­ing mun­dane and rou­tine work. This will al­low lo­gis­tics work­forces to fo­cus on more mean­ing­ful and im­pact­ful work.


There are many rea­sons to be­lieve that now is the best time for the lo­gis­tics in­dus­try to em­brace AI. Never be­fore has this ma­tur­ing tech­nol­ogy been so ac­ces­si­ble and af­ford­able. This has al­ready made nar­row forms of AI ubiq­ui­tous in the con­sumer realm; en­ter­prise and in­dus­trial sec­tors are soon to fol­low.

In lo­gis­tics, the net­work-based na­ture of the in­dus­try pro­vides a nat­u­ral frame­work for im­ple­ment­ing and scal­ing AI, am­pli­fy­ing the hu­man com­po­nents of highly or­gan­ised global

sup­ply chains. Fur­ther­more, com­pa­nies de­cid­ing not to adopt AI run the risk of ob­so­les­cence in the long term, as com­peti­tors seize and ef­fec­tively use AI in their busi­ness to­day.

Re­searchers at IBM es­ti­mate only 10 per cent of cur­rent sys­tems, data, and in­ter­ac­tions in­clude el­e­ments of AI anal­y­sis and re­sults. How­ever, the re­turns on AI in­vest­ments are al­ready im­prov­ing; rel­a­tively mod­er­ate out­lay is gen­er­at­ing a much larger re­turn than ever be­fore.

But as com­plex­ity grows – with more un­struc­tured data, more so­phis­ti­cated learn­ing al­go­rithms and tech­niques, and more high-level de­ci­sion-mak­ing tasks – the cu­mu­la­tive na­ture of AI means that AI anal­y­sis and re­sults will im­prove even fur­ther.

There is an­other in­di­ca­tor that now is a good time for AI to flour­ish – this is the state of its adop­tion in the world to­day. In­no­va­tions oc­cur first and be­come main­stream in the con­sumer world. Once a tip­ping point is reached, these in­no­va­tions work their way into com­mer­cial en­ter­prises and ul­ti­mately into in­dus­trial com­pa­nies.

AI is stretch­ing be­yond con­sumer ubiq­uity and into cus­tomer-fo­cused com­mer­cial ven­tures. Even­tu­ally, once the value of AI is proven in the com­mer­cial con­text, it will ar­rive in the in­dus­trial set­ting. The spe­cific tim­ing of these transitions is im­pos­si­ble to pre­dict, but the fact that AI is now deeply em­bed­ded in con­sumer mar­kets and is ex­pe­ri­enc­ing ex­plo­sive growth in cus­tomer-fac­ing com­mer­cial ar­eas clearly in­di­cates the use of AI in in­dus­trial sec­tors such as lo­gis­tics is quickly ap­proach­ing.

Lo­gis­tics com­pa­nies are uniquely po­si­tioned to ben­e­fit by ap­ply­ing AI in al­most all as­pects of the sup­ply chain. One of the most un­der­utilised as­sets in the in­dus­try is the high vol­ume of data that sup­ply chains gen­er­ate on a daily ba­sis.

This data is both struc­tured and un­struc­tured, and AI will en­able lo­gis­tics com­pa­nies to ex­ploit it. In ad­di­tion, as many lo­gis­tics com­pa­nies around the world em­brace dig­i­tal trans­for­ma­tion, tran­si­tion­ing away from legacy en­ter­prise re­source plan­ning sys­tems to ad­vanced an­a­lyt­ics, in­creased au­to­ma­tion, and hard­ware and soft­ware ro­bot­ics, and mo­bile com­put­ing, the next ob­vi­ous step in the in­creas­ingly dig­i­tal sup­ply chain is to ap­ply AI. Fur­ther­more, lo­gis­tics com­pa­nies de­pend on net­works – both phys­i­cal and in­creas­ingly dig­i­tal – which must func­tion har­mo­niously amid high vol­umes, low mar­gins, lean as­set al­lo­ca­tion, and time-sen­si­tive dead­lines. AI of­fers lo­gis­tics com­pa­nies the abil­ity to op­ti­mise net­work or­ches­tra­tion to de­grees of ef­fi­ciency that can­not be achieved with hu­man think­ing alone.

AI can help the lo­gis­tics in­dus­try to re­de­fine to­day’s be­hav­iours and prac­tices, tak­ing op­er­a­tions from re­ac­tive to proac­tive, plan­ning from fore­cast to pre­dic­tion, pro­cesses from man­ual to au­ton­o­mous, and ser­vices from stan­dard­ised to per­son­alised.


In an in­creas­ingly com­plex and com­pet­i­tive busi­ness world, com­pa­nies that op­er­ate global sup­ply chains are un­der un­prece­dented pres­sure to de­liver higher ser­vice lev­els at flat or even lower costs. At the same time, in­ter­nal func­tions of global cor­po­ra­tions, such as ac­count­ing, fi­nance, hu­man re­sources, le­gal, and in­for­ma­tion tech­nol­ogy are plagued by large amounts of de­tail-ori­ented,

“AI tech­nolo­gies like nat­u­ral lan­guage pro­cess­ing can ex­tract crit­i­cal in­for­ma­tion.”

repet­i­tive tasks. Here, AI presents a sig­nif­i­cant op­por­tu­nity to save time, re­duce costs and in­crease pro­duc­tiv­ity and ac­cu­racy with cog­ni­tive au­to­ma­tion.

Cog­ni­tive au­to­ma­tion refers to in­tel­li­gent busi­ness process au­to­ma­tion us­ing a com­bi­na­tion of AI and ro­botic process au­to­ma­tion (RPA). This is the re­place­ment of cler­i­cal la­bor us­ing soft­ware ro­bots that can be in­te­grated into ex­ist­ing busi­ness ap­pli­ca­tions and IT sys­tems.

RPA is not equiv­a­lent to AI; where AI is able to learn and ex­tract in­sights from un­struc­tured data, RPA is able to ex­e­cute rule-based work­streams given well-struc­tured in­puts on be­half of hu­man work­ers, and can­not learn be­yond its ini­tial pro­gram­ming.


Lo­gis­tics ser­vice providers of­ten rely on vast num­bers of third par­ties, in­clud­ing com­mon car­ri­ers, sub­con­tracted staff, char­ter air­lines, and other third-party ven­dors to op­er­ate core func­tions of their busi­ness. This puts an in­creased bur­den on lo­gis­tics ac­count­ing teams to process mil­lions of in­voices an­nu­ally from thou­sands of ven­dors, part­ners, or providers.

Here, AI tech­nolo­gies like nat­u­ral lan­guage pro­cess­ing can ex­tract crit­i­cal in­for­ma­tion, such as billing amounts, ac­count in­for­ma­tion, dates, ad­dresses, and par­ties in­volved from the sea of un­struc­tured in­voice forms re­ceived by the com­pany.

Once the data is well clas­si­fied, an RPA bot can take it and in­put it into ex­ist­ing ac­count­ing soft­ware to gen­er­ate an or­der, ex­e­cute pay­ment, and send the cus­tomer a con­fir­ma­tion email, all with­out hu­man in­ter­ven­tion.

Con­sul­tancy firm Ernst & Young (EY) is ap­ply­ing a sim­i­lar ap­proach for detection of fraud­u­lent in­voices. Us­ing ma­chine learn­ing to thor­oughly clas­sify in­voices from in­ter­na­tional par­ties and iden­tify anom­alies for ex­pert re­view helps EY com­ply with sanc­tions, anti-bribery reg­u­la­tions, and other as­pects of the US For­eign

Cor­rupt Prac­ticesAct. EY’s fraud detection sys­tem achieves 97 per cent ac­cu­racy and has been rolled out to more than 50 com­pa­nies. Sim­i­lar logic can be ap­plied to any busi­ness process with high-fre­quency repet­i­tive tasks.


Global lo­gis­tics and sup­ply chain op­er­a­tors typ­i­cally man­age large fleets of ve­hi­cles and net­works of fa­cil­i­ties world­wide.

Ger­man real es­tate soft­ware-as-a-ser­vice (SaaS) firm Lev­er­ton uses AI on its plat­form of the same name to sim­plify the pro­cess­ing and man­age­ment of real es­tate con­tracts for busi­nesses. The sys­tem uses nat­u­ral lan­guage pro­cess­ing to clas­sify any con­trac­tual clauses, pol­icy-rel­e­vant sec­tions, and sig­na­ture por­tions.

Paired with a hu­man-in-the-loop to re­view these find­ings, con­tracts writ­ten in com­plex le­gal lan­guage – of­ten sev­eral hun­dred pages in length – can be pro­cessed in a frac­tion of the time it would take a team of hu­man ex­perts.

Keep­ing cus­tomer in­for­ma­tion up to date is a chal­lenge for large en­ter­prises; up to 25 per cent of all phone num­bers and email ad­dresses stored in dig­i­tal con­tact ap­pli­ca­tions are no longer in use. In the lo­gis­tics in­dus­try, keep­ing

ad­dress in­for­ma­tion com­plete and cur­rent is crit­i­cal for suc­cess­ful de­liv­ery of ship­ments.

Of­ten, large teams of data an­a­lysts are tasked with cus­tomer re­la­tion­ship man­age­ment (CRM) clean-up ac­tiv­i­ties, elim­i­nat­ing du­pli­cate en­tries, stan­dar­d­is­ing data for­mats, and re­mov­ing out­dated con­tacts.

Amer­i­can startup Cir­cleBack has de­vel­oped an AI en­gine to help man­age con­tact in­for­ma­tion, con­tin­u­ally pro­cess­ing bil­lions of data points to de­ter­mine whether con­tact in­for­ma­tion is ac­cu­rate and up to date. AI tools trained in in­put man­age­ment can use nat­u­ral lan­guage pro­cess­ing to do some pre-pro­cess­ing of cus­tomer ad­dress in­for­ma­tion to en­sure com­plete­ness, cor­rect­ness, and con­sis­tency with global and re­gional ad­dress for­mats.


AI also stands to greatly ben­e­fit the phys­i­cal de­mands of work­ing in mod­ern lo­gis­tics.

The use of AI-en­abled ro­bot­ics, com­puter vi­sion sys­tems, con­ver­sa­tional in­ter­faces, and au­ton­o­mous ve­hi­cles is the phys­i­cal em­bod­i­ment of AI in lo­gis­tics op­er­a­tions, wel­com­ing in a new class of tools to aug­ment the ca­pa­bil­i­ties of to­day’s work­force. In­tel­li­gent ro­botic sort­ing is the ef­fec­tive high-speed sort­ing of let­ters, parcels, and even pal­letised ship­ments – one of the most crit­i­cal ac­tiv­i­ties of mod­ern par­cel and ex­press op­er­a­tors. Ev­ery day, mil­lions of ship­ments are sorted with a so­phis­ti­cated ar­ray of con­vey­ors, scan­ning in­fra­struc­ture, man­ual han­dling equip­ment, and per­son­nel. The lo­gis­tics in­dus­try can draw on AI-driven ro­bot­ics in­no­va­tions from the re­cy­cling in­dus­try.

Fin­nish com­pany SenRobotics has been de­vel­op­ing in­tel­li­gent ro­botic waste sort­ing sys­tems since 2011. The com­pany’s SRR2 ro­botic sys­tem uses a com­bi­na­tion of com­puter vi­sion and ma­chine learn­ing al­go­rithms em­bed­ded in off-the-shelf ro­botic arms in a syn­chro­nised way to sort and pick re­cy­clables from mov­ing con­veyor belts. The AI en­gine in­gests real-time data from three dif­fer­ent cam­eras and sen­sor types, and is trained to iden­tify a wide va­ri­ety of food and bev­er­age car­tons by recog­nis­ing lo­gos, la­bels, and 3D forms.

The re­sult is a sys­tem con­sist­ing of two AI-pow­ered ro­botic arms that can sort un­struc­tured re­cy­clables on a mov­ing con­veyor belt at a rate of 4,000 items per hour with high de­grees of pre­ci­sion. This sug­gests a use­ful AI ap­pli­ca­tion in lo­gis­tics. Sim­i­lar sort­ing ca­pa­bil­i­ties could the­o­ret­i­cally be ap­plied to par­cel and let­ter-sized ship­ments to re­duce hu­man ef­fort and er­ror rates.

Au­ton­o­mous guided ve­hi­cles (AGVs) are al­ready start­ing to play a big­ger role in lo­gis­tics op­er­a­tions. Within any given lo­gis­tics op­er­a­tion, it is typ­i­cal to see mul­ti­ple peo­ple op­er­at­ing ma­te­rial han­dling equip­ment such

“AI also stands to greatly ben­e­fit the phys­i­cal de­mands of work­ing in mod­ern lo­gis­tics.”

as fork­lifts, pal­let jacks, wheeled totes, and even tug­ging cars to move goods be­tween lo­ca­tions or ves­sels. To re­duce this, com­pa­nies are be­gin­ning to use non-in­dus­trial, col­lab­o­ra­tive ro­bot­ics, in­clud­ing AGVs. AI is a key part of this.

GreyOrange, a Sin­ga­pore-founded au­to­ma­tion and ro­bot­ics com­pany that de­vel­ops self-nav­i­gat­ing AGVs, re­cently also launched GreyMat­ter, its next-gen­er­a­tion soft­ware plat­form.

One of the com­pany’s launch part­ners for both in­no­va­tions is Ni­tori, a Ja­panese fur­ni­ture and home fur­nish­ings chain. As the name sug­gests, GreyMat­ter makes use of AI to al­low real-time col­lab­o­ra­tion be­tween AGVs, en­abling op­ti­mised nav­i­ga­tion path plan­ning, zon­ing, and speeds, as well as self-learn­ing to im­prove AGV ca­pa­bil­i­ties over time. When given or­ders to ful­fil, the AGVs and the plat­form are aware of each item that is be­ing trans­ported and the routes that are taken to re­trieve and de­liver these items. Ni­tori is us­ing this valu­able data to achieve the most ef­fi­cient han­dling routes and pre­dict prod­uct pop­u­lar­ity and sea­sonal trends – self-op­ti­mis­ing op­er­a­tions that en­sure ever-short­en­ing ful­fil­ment times as well as real-time vis­i­bil­ity of prod­uct de­mand.


AI-Pow­ered Vis­ual In­spec­tion is an­other high-po­ten­tial area for AI in the lo­gis­tics op­er­a­tional en­vi­ron­ment. Ad­vances in com­puter vi­sion are al­low­ing us to see and un­der­stand the world in new ways, and lo­gis­tics op­er­a­tions are no ex­cep­tion.

IBM Wat­son, the com­puter sys­tem ca­pa­ble of an­swer­ing ques­tions posed in or­di­nary lan­guage, is us­ing its cog­ni­tive vis­ual recognition ca­pa­bil­i­ties to do main­te­nance of phys­i­cal as­sets with AI-driven vis­ual in­spec­tion.

In in­dus­trial sec­tors like lo­gis­tics, dam­age and wear to op­er­a­tional as­sets over time are sim­ply in­her­ent.

Us­ing a cam­era bridge to pho­to­graph cargo train wag­ons, IBM Wat­son was re­cently able to suc­cess­fully iden­tify dam­age, clas­sify the dam­age type, and de­ter­mine the ap­pro­pri­ate cor­rec­tive ac­tion to re­pair these as­sets. First, cam­eras were in­stalled along train tracks to gather im­ages of train wag­ons as they drove by. The im­ages were then au­to­mat­i­cally up­loaded to an IBM Wat­son im­age store where AI im­age clas­si­fiers iden­ti­fied dam­aged wagon com­po­nents.

The AI clas­si­fiers were trained on where to look for wagon com­po­nents in a given im­age and how to suc­cess­fully recog­nise wagon parts and then clas­sify them into seven dam­age types. As more data was gath­ered and pro­cessed, Wat­son’s vis­ual recognition ca­pa­bil­i­ties im­proved to an ac­cu­racy rate of over 90 per cent in just a short pe­riod of time. The anom­alies and dam­ages dis­cov­ered by Wat­son were sent to a work­place dash­board man­aged by main­te­nance teams. This model and process can loosely be ap­plied to other types of lo­gis­tics as­set in­clud­ing, but not lim­ited to air­craft, ve­hi­cles, and ocean ves­sels. Com­puter vi­sion in­ven­tory man­age­ment and ex­e­cu­tion are be­com­ing re­al­ity to­day in the re­tail in­dus­try.

French startup Qopius is de­vel­op­ing

com­puter vi­sion-based AI to mea­sure shelf per­for­mance, track prod­ucts, and im­prove re­tail store ex­e­cu­tion. Us­ing deep learn­ing and fine-grained im­age recognition, Qopius is able to ex­tract char­ac­ter­is­tics of items such as brand, la­bels, lo­gos, price tags, as well as shelf con­di­tion – for ex­am­ple, out of stock, share of shelf, and on-shelf avail­abil­ity. In ware­house in­ven­tory man­age­ment, sim­i­lar use of com­puter vi­sion AI of­fers po­ten­tial for real-time in­ven­tory man­age­ment at the in­di­vid­ual piece and stock-keep­ing unit level.

Cana­dian startup Twen­tyBN is work­ing on deep-learn­ing AI that is able to de­ci­pher com­plex hu­man be­hav­iour in video streams. Pre­vi­ous ap­pli­ca­tions of its tech­nol­ogy in­clude au­ton­o­mous detection from video feeds alone of things like an elderly per­son fall­ing, ag­gres­sive be­hav­iour on pub­lic trans­port, and shoplift­ing in stores. Con­sid­er­ing that many ware­houses to­day are equipped with surveil­lance cam­eras for safety pur­poses, this type of AI tech­nol­ogy can be used to op­ti­mise per­for­mance (by de­tect­ing, for ex­am­ple, suc­cess­ful pick and pack tasks) and in­crease op­er­a­tional safety (for ex­am­ple, with in­stant alert­ing of ac­ci­dents in­volv­ing work­ers).

Con­ver­sa­tional in­ter­faces are be­com­ing in­creas­ingly com­mon in the con­sumer world. Voice-based pick­ing has been around in sup­ply chain op­er­a­tions since the 1990s, but re­cent break­throughs in nat­u­ral lan­guage pro­cess­ing are bring­ing new con­ver­sa­tional ca­pa­bil­i­ties to the sup­ply chain.

Amer­i­can startup AVRL is en­abling tra­di­tional in­dus­trial IT plat­forms with con­ver­sa­tional ca­pa­bil­i­ties via pro­pri­etary, nat­u­ral lan­guage AI. Be­fore the par­al­lel ad­vance­ments of AI and speech recognition tech­nolo­gies, voice-en­abled tools in the sup­ply chain were static; they were lim­ited to key­words and au­di­ble menus, and op­er­ated with fixed com­mands. Fur­ther­more, these sys­tems were lim­ited in terms of in­ter­ac­tion, sup­port­ing only a num­ber of lan­guages, ac­cents and di­alects. As a re­sult, hu­mans had to rely on rel­a­tively scripted re­sponses to op­er­ate these rigid voice sys­tems.


While there are myr­iad fac­tors in­flu­enc­ing the devel­op­ment, ac­cep­tance, and dis­tri­bu­tion of fully au­ton­o­mous trans­porta­tion, this sec­tion ex­am­ines how AI is con­tribut­ing to the progress of au­ton­o­mous ve­hi­cles.

For fully au­ton­o­mous ve­hi­cles to be­come widely ac­cepted, they need to sig­nif­i­cantly out­per­form hu­man driv­ing ca­pa­bil­i­ties. This be­gins by en­abling the ve­hi­cle to per­ceive and pre­dict changes in its en­vi­ron­ment – some­thing that is sim­ply im­pos­si­ble with­out AI.

Au­ton­omy to­day re­lies on a suite of sens­ing tech­nolo­gies that work to­gether to pro­duce a high-def­i­ni­tion three-di­men­sional map of the ve­hi­cle’s en­vi­ron­ment. Deep-learn­ing al­go­rithms on board the ve­hi­cle process this live stream of en­vi­ron­men­tal data to iden­tify ob­sta­cles and other cars, in­ter­pret road signs, street mark­ings, and traf­fic sig­nals, and com­ply with speed lim­its and traf­fic laws. Since there is no pos­si­ble way to hard-pro­gram an au­ton­o­mous ve­hi­cle to re­act to ev­ery pos­si­ble driv­ing sce­nario in the real world, de­vel­op­ers must turn to the con­tin­u­ous knowl­edge ac­qui­si­tion of deep learn­ing. This way they can de­velop au­ton­o­mous ve­hi­cles that con­stantly im­prove their ca­pa­bil­i­ties as they are in­tro­duced to new sur­round­ings.

Tra­di­tional auto in­dus­try play­ers such as BMW, Daim­ler, Ford, Toy­ota, and VW have em­braced AI as a crit­i­cal com­po­nent in their au­ton­o­mous ve­hi­cle devel­op­ment jour­neys. More fa­mously, newer en­trants like Google, Tesla, and Waymo have de­vel­oped their own au­ton­o­mous ve­hi­cles us­ing pro­pri­etary AI and man­u­fac­tur­ing

“Au­ton­o­mous fleets will even­tu­ally be used in all as­pects of the sup­ply chain from end to end.”

tech­niques. On the other hand, au­to­mo­tive sup­pli­ers such as Bosch, Mo­bil­eye, Nvidia, Quan­ergy, and SF are mak­ing com­po­nents in­clud­ing sen­sors, al­go­rithms, and data avail­able to sup­port fur­ther devel­op­ment of au­ton­o­mous ve­hi­cles. In ad­di­tion, mo­bil­ity plat­form com­pa­nies Lyft and Uber are part­ner­ing with es­tab­lished au­to­mo­tive com­pa­nies to of­fer on-de­mand rides au­tonomously.

Con­ve­nience, cost re­duc­tion, and in­creased ef­fi­ciency in the form of lower emis­sions and fewer ac­ci­dents are the pri­mary driv­ers for au­ton­o­mous ve­hi­cle use. Thanks to the fall­ing cost of com­po­nents, in­creas­ing per­for­mance of deep learn­ing al­go­rithms, and the grow­ing col­lec­tive body of trans­porta­tion in­dus­try knowl­edge on the topic of au­ton­o­mous driv­ing, the sup­port­ing tech­nol­ogy is de­vel­op­ing rapidly.

How­ever, full im­ple­men­ta­tion – in other words, a ve­hi­cle with­out a driver in the le­gal sense – will ne­ces­si­tate sig­nif­i­cant reg­u­la­tory changes in any coun­try and this will take some time.


Au­ton­o­mous fleets will even­tu­ally be used in all as­pects of the sup­ply chain from end to end. Early signs of this can be seen in in­tra-lo­gis­tics, line-haul truck­ing, and last-mile de­liv­ery.

Truck pla­toon­ing refers to the in­tel­li­gent car­a­van­ning of groups of semi-trucks. With ma­chine-to-ma­chine com­mu­ni­ca­tions and col­lab­o­ra­tive as­sisted cruise-con­trol tech­nol­ogy, be­tween two to five semi-trucks can fol­low each other and au­tonomously syn­chro­nise ac­cel­er­a­tion, steer­ing, brak­ing and fol­low­ing dis­tance. The pla­toon is con­trolled by a hu­man driver op­er­at­ing the lead truck, with a backup driver in each fol­low­ing truck if needed.

The British Trans­porta­tion Re­search Lab­o­ra­tory, to­gether with DHL and DAF Trucks, will pi­lot a pla­toon­ing project on UK mo­tor­ways in 2019. This is one of many re­cent and planned pla­toon­ing tri­als across the truck­ing in­dus­try.

Au­ton­o­mous ve­hi­cles and trucks to­day ac­tu­ally han­dle free­way driv­ing rel­a­tively well, and each pla­toon­ing trial or au­ton­o­mous free­way mile driven by semi-trucks adds new data for all trucks to im­prove their au­ton­o­mous driv­ing ca­pa­bil­ity.

Over time, this ca­pa­bil­ity will con­tinue to get bet­ter un­til fully au­ton­o­mous trucks be­come a re­al­ity. Ex­perts es­ti­mate that free­way au­ton­omy in­tel­li­gence is largely com­plete; the main chal­lenge is now the miles on smaller streets be­tween free­ways and des­ti­na­tions.

In re­cent years, there has been ex­ten­sive ex­per­i­men­ta­tion with last-mile de­liv­ery with the aim of re­duc­ing cost and com­plex­ity.

One method in par­tic­u­lar has cap­tured con­sumer imag­i­na­tion and me­dia head­lines – de­liv­ery via au­ton­o­mous un­manned aerial ve­hi­cles – but reg­u­la­tory frame­works and suf­fi­ciently dense road de­liv­ery net­works limit the com­mer­cial vi­a­bil­ity of this to ru­ral ar­eas at best. A more prac­ti­cal ap­pli­ca­tion of au­ton­omy in the last mile comes in the form of au­ton­o­mous un­manned ground ve­hi­cles (UGVs) that op­er­ate fully au­tonomously or in col­lab­o­ra­tion with a de­liv­ery per­son.

Amer­i­can startup Robby Tech­nolo­gies is de­vel­op­ing Robby 2, an au­ton­o­mous un­manned ground ve­hi­cle with ad­vanced AI for nav­i­ga­tion and in­ter­ac­tion ca­pa­bil­i­ties.

Given the dy­namic com­plex­ity of nav­i­gat­ing side­walks, pedes­tri­ans, road and rail cross­ings, Robby’s on-board AI is called upon to con­tin­u­ously sense and re­act to novel sit­u­a­tions and be­come smarter with use. Em­bed­ded con­ver­sa­tional AI helps im­prove in­ter­ac­tion be­tween hu­mans and the robot; if a per­son blocks Robby’s path, a soft voice po­litely asks “ex­cuse me” and will even say “thank you” when a per­son makes room for the robot. With grow­ing de­liv­ery de­mand in dense ar­eas, au­ton­o­mous ro­bots like Robby can help ex­ist­ing last-mile de­liv­ery fleets man­age in­creased vol­umes at lower cost.

Top: A full AI learn­ing cy­cle

Be­low: IBM global travel and trans­porta­tion rail leader Keith Dierkx

Above: DHL Se­nior vice pres­i­dent strat­egy, mar­ket­ing and in­no­va­tion Matthias Heut­ger

Be­low: See­ing, speak­ing and think­ing lo­gis­tics op­er­a­tions

Be­low: Deep learn­ing in ac­tion

Newspapers in English

Newspapers from Australia

© PressReader. All rights reserved.