Since UPS ratified their contract with the Teamsters union, social media has been awash with memes about UPS drivers making $170k a year.
I won’t get into the breakdown of salary vs benefits, but needless to say that with the average UPS driver set to earn up to $170k a year by 2028, the already most expensive part of parcel delivery (the last mile) is set to continue to outpace inflation.
With US Online spend set to grow from $1.148t in 2023 to $1.73t by 2027 according to eMarketer and corresponding US package volume increasing from 22 billion packages in 2023 to 26.9 billion by 2027 according to Pitney Bowes, shippers are desperately seeking ways to lower final mile delivery spend.
Reason the last mile costs continue to soar is that unlike in the 1st and middle mile, where warehouse and sortation automation solutions are already mature and being rapidly deployed both in new and existing facilities resulting in increased scalability, lower labor costs and longer operating hours, to date there has been virtually no automation in the last mile. The status quo is that 100% of last mile deliveries in the US are still performed by human delivery drivers.
In this blog post, I’m going to take a forward looking (10 year) review of last mile automation options that may provide viable alternatives to the human delivery driver and allow the eCommerce parcel industry to put into reverse the spiraling cost of delivering packages to the consumer’s doorstep.
Drones – The Hype Cycle That Never Was Nor Never Will Be
Drone delivery was all the hype until it wasn’t anymore. 10 years ago, Jeff Bezos was full of confidence that 85% of Amazon packages under 5lbs would be delivered by his company’s drones by 2018. We were led to believe our packages would fall from the sky like presents from Santa’s sleigh, so what happened ?
In hindsight, while drones will undoubtedly have a role to play in niche consumer delivery scenarios, it now seems rather obvious that last mile delivery by drone was nothing more than a billionaire’s fantasy. The challenges of turning this fantasy to reality kept on chalking up. Just some of the issues that surfaced were:
- Battery Life. Even with recent (and future anticipated) advancement in battery technology, drones are still only capable of flying ~25 minutes while carrying a relatively modest payload of less than 5lbs. This means in reality they can only perform 1 delivery at a time before having to return to the DC to pick up the next package. Assuming no battery recharging time (i.e. replacing the battery on each trip) and a delivery operational window of 8am – 9pm, that’s a max of ~30 deliveries per-drone, per-day, which is far less than the 150+ deliveries an equivalent UPS / FedEx driver typically performs on a shift.
The number of drones required to replace existing driver based deliveries and scale for growth over the next 10 years would number into tens of millions.
- Weather. Drones can’t fly in heavy rain, snow, extreme low temperatures or high winds. This makes them akin to the renewable energy problem. On the days when the drones can’t fly, there needs to be a conventional backup. This simply isn’t economical making this one of the biggest roadblocks to drone delivery reality.
- Landing. In almost all the pilot programs operated by Amazon and others, the consumer had to place a QR code landing mat in their yard that the drones cameras could identify and navigate to. America’s backyards present an almost inhospitable environment for a drone trying to safely land. Trees, swimming pools, power lines, hedges, fences and dogs (all dogs HATE drones and will bark, jump up and attack any drone attempting a landing in their yard) all make landing a drone in a yard a treacherous process.
- Dropping. Because landing is almost a non-starter, most drone pilots experimented with dropping packages using a parachute or lowering the package safely to the ground via a retractable cable like a search and rescue helicopter. On paper this might seem easier than landing a drone, but parachutes only work for non breakable goods and cables have a high risk of getting tangled up in a tree or other obstacles. “Dropping” a package requires a roomy, controlled, obstruction free area in the consumers yard.
- The FAA. The FAA restricts drone operations in large areas surrounding both commercial and military airports. This means that large parts of the population density of almost every US city are currently and likely will remain off limits, leaving drone delivery scope to the outer suburbs and rural areas.
- Home Type. Most drone delivery trials were focused on making deliveries to single family homes (with a yard). Drones are basically a non-starter for condos, apartment buildings, inner city semi detached homes and any home without a (big) yard.
So if all these problems are seemingly insurmountable, why has Amazon alone invested over 10 years and $2b developing their Drone program? Well, it might seem rather obvious in hindsight, but for a few years the media and almost everyone else built up a frenzy that attracted big money from VC firms and others. It wasn’t just Amazon that thought Drones were the answer, DHL after investing in its own ParcelCopter drone announced back in 2021 they were shutting down the program.
In Gartner’s August 2023, Hype Cycle for Mobile Robots and Drones, 2023 the analyst firm puts Drones (or what they call “Autonomous Mobile Transport Robots”) at the very bottom of the trough of disillusionment and calls out (correctly) that:
“delivery drones for consumer deliveries are the ultimate aim to create scale and reduce costs. However, for at least the next five years, we expect use cases to be focused on public services (e.g., medical and relief deliveries) and B2B services, where small-scale deployments can be profitable and legislation is less restrictive.”
Amen – Drones will have a role to play in B2B deliveries where the landing zone is in a designated and controlled area and may play a role in niche or rural consumer deliveries, but unionized UPS delivery drivers don’t need to stay awake at night worrying about losing their job to a drone.
So if Drones aren’t the answer, what is?
That’s the million dollar question. If flying packages to the consumers doorstep isn’t the answer, then what ground based automation options exist ? Let’s take a look:
#1 – Autonomous wheeled robots
Across Europe, the US and in China there have been a plethora of VC funded startups and established carrier trials of wheeled delivery robots. A few notable examples include JD.com, StarShip, FedEx’s Roxo (shut down in 2022) and DPD. But just like drones don’t work well delivering packages into America’s backyard, neither it turns out do wheeled robots. Like their drone counterparts, the wheeled robot has some major shortcomings that simply make them almost useless for consumer parcel delivery. Here’s why:
- The world isn’t flat. These robots can only drive on flat surfaces, making them either very bad at or at worst completely useless at navigating curbs, steps, gravel surfaces, etc. My 13 year old niece lives in Cambridge UK where StarShip operates a food delivery service. Multiple times a day she plays the good citizen helping out a StarShip that has fallen on its side trying to navigate a curb, or has got stuck in some kind of situation its autonomous navigation system couldn’t handle. Each time she helps the Starship on its way, the robot politely thanks her in a robotic voice – but having armies of teenagers helping the drones back onto their feet isn’t a scalable answer.
- Obstacles are everywhere. These wheeled devices can’t open gates, ring doorbells or climb stairs making them fairly useless as navigating the majority of front yards. Their deployment use cases are limited to unobstructed, flat driveway single detached homes.
- Dogs. Oh the dog thing again. I’m sorry but EVERY DOG is going to bark and go completely mental at an autonomous drone driving up its driveway. Anyone with kids, a remote control car and a dog will understand what I’m talking about. Yes dogs have always hated the postman and the UPS driver, but they really really hate remote controlled cars (which is what a wheeled robot is to them).
- Theft and vandalism. It turns out, cute little robots rolling up and down the street are great targets for theft and vandalism. In LA operators are increasingly seeing their robots vandalized, stolen or the secure package trunk forced open to steal the contents inside. In fact a whole underground movement has emerged against these robots with attacks being uploaded and shared on social channels. The little robots can’t defend themselves and at best can make a noise like a car alarm – but they are easy targets. Vandalism and theft are considered petty crimes in most jurisdictions whereas stealing a package from a UPS driver would be considered assault. The latter most likely comes with a prison sentence, the former doesn’t.
- They don’t have limbs. While the obstacles listed above may seem enough to kill the wheeled robot on their own, the single biggest issue with wheeled drones is they were designed to serve a fast food or grocery delivery use case that assumes that the consumer will be at home, willing and ready to come outside and retrieve their food from the “secure” belly of the robot. Guess what? This doesn’t work for parcel delivery – the consumer ain’t at home, or they ain’t coming outside to retrieve their online order from the robots belly. They expect the robot to leave their package on the doorstep, or put it in their mailbox. Well these wheeled robots don’t have arms or hands, so they have no way to securely and safely transfer the contents of their belly onto the customers doorstep.
#2 – Autonomous Canine Robots
Okay, so if wheeled robots aren’t the answer, what about a robot with 4 legs and a robotic arm instead?
Let me introduce you to Spot the Robodog. Spot is a commercially successful project from Boston Dynamics. He’s got a pedigree that on paper outshines his airborne and wheeled counterparts:
- Battery Life – Spot has up to 90 minutes of runtime, with an ability to recharge to 80% in 50 minutes (not great – but the batteries can be swapped), meaning that Spot could in theory do a full days delivery shift simply by replacing the batteries every hour or so.
- Payload – Spot Can carry packages up to 24 lbs in his mouth (they call it the Arm) meaning he could deliver the majority of small parcels
- Agility – Spot can open gates and doors, climb steps and jump over obstacles (something his wheeled counterparts can’t do) meaning he can navigate even some of the most complex front yard environments and could even get into apartment buildings and climb stairs to make deliveries above (or below) ground level.
- Autonomous – Spot can navigate around previously encountered surroundings using visual and lidar inputs, meaning he can be trained to navigate around commonly encountered obstacles like a car parked on a driveway.
So, at first glance Spot seems well placed to take on the role of replacing the UPS delivery driver. Not so fast, it turns out that Spot also has some major shortcomings that make him not ready to roam America’s streets just yet:
- Weather. Spot is kinda a dry weather creature. It turns out he’s not certified to work in the rain or snow (yet) – his hydraulic actuators and electronics aren’t yet waterproof. Back to the same problem that drones and wheeled robots have – they are useless in inclement weather conditions.
- He’s Only Got One Arm. Spot can open doors, gates etc with his arm, but if that arm is carrying a parcel, then this ability is eradicated. Spot needs two arms (like a human) to be able to carry a parcel and open doors / gates at the same time.
- His mouth isn’t big enough. Spot has a mouth (or maybe it’s a hand ?) on the end of his arm, but it only opens to a max aperture of ~6 inches, meaning he’s not going to be able to carry bulkier packages like a shoe box.
- He has no feelings. His mouth / hand can clamp but the “Arm” actuator doesn’t have any pressure sensitivity so he can’t tell is he’s applying enough or too much pressure to the parcel he’s carrying (oops – lots of crushed or dropped boxes resulting in insurance claims to the carrier)
- He needs a pre-programed route. Although spot has sensors that allow him to navigate obstacles, open doors etc., he was designed to do repetitive tasks like walk around an industrial complex and take sensor readings. Thus he needs to be taught a route (by an operator) and can only deviate from the route by a few feet should he encounter an unexpected obstacle. He doesn’t have an AI based vision and navigation system (like Telsa autopilot) that would allow him to be given a delivery address and navigate down an unfamiliar street to the address, recognise it reading a house number and / or using GPS.
Again Gartner has these “Smart Robots” in the right place in the Hype Cycle, in the “Peak of Inflated Expectations” phase. So although Spot clearly isn’t ready for the job what other options are there?
#3 – Autonomous Humanoid Robots
So now we are really entering the sci-fi era. Or are we? If Elon Musk has anything to say about it, then the Humanoid Robot is much closer to reality than any of us currently comprehend. Just take a look at this video recently released by Tesla’s AI team.
Yes there have been other Humanoid Robots before, namely Boston Dynamics Atlas. Most of us have probably seen videos on YouTube of these robots doing Parkour or lifting heavy objects, but Atlas (like Spot), was designed to be used in controlled industrial environments, not in public spaces.
This is where the Tesla Bot is different. It is being designed from the ground up as a consumer device. A home assistant that can vacuum, fold laundry and cut the grass. According to Gartner, Personal robots are still in the nascent “Innovation Trigger” phase, but let’s take a closer look to see if the Tesla Bot could one day be the answer for replacing the UPS Driver?
- Neural Network AI based learning. This is the BIG differentiator vs what has gone before. For those not familiar with the latest incarnation (v12) of Tesla’s self-driving technology, it’s completely run on a neural network, meaning that there are no lines of code, algorithms or logic programs telling the robot how to do something. It learns 100% from being taught how to do a task, then it can do that task over and over again and gets better and better at it (just like a human). Tesla has taken the same Neural Network technology from their self driving car program and transplanted it into their humanoid bot. The significance of this is perhaps hard to grasp if you don’t follow along with what Tesla have been doing lately, but let me try and break it down.
The carrier simply needs to teach the bot what to do. This could actually be as easy as equipping existing drivers with body cameras to capture millions of hours of real life delivery scenarios. Just like Tesla’s autopilot the footage would be ingested by Telsa’s Dojo AI training computers and the neural network does the rest. It simply learns about all the different types of packages, homes and apartments and how to handle each one of them. It can learn how to open gates, ring buzzers, ignore barking dogs and stand back up after it slips on ice or snow. The list goes on and on to infinity. The point is that every possible scenario encountered by a human delivery driver is captured and the neural network trains itself on how to respond based on the aggregate footage from the body cameras. As the bots themselves are unleashed into the wild, all of their footage continues to feed into the network, continuously providing training material that makes the robots able to understand and tackle every single edge case scenario.
- Agility. The Tesla Bot is a humanoid. It has the same range of motion, balance and agility as a human, but is much much stronger, meaning it can potentially carry heavier packages than humans can today. Anything the human delivery driver can do, the humanoid robot can (in theory) do. Open a gate, jump a fence, climb up a snow bank, open a door, climb a staircase – the list goes on and on.
Regardless if Wheeled Robots, Canine Robots or Humanoid Robots can evolve past the technical hurdles outlined above, there are still some other major hurdles that need to be addressed before we will see robot deliveries become anything more than another billionaires fantasy:
- Consumer acceptance. In theory consumers shouldn’t care how their parcel gets to their doorstep as long as it gets there on time and undamaged. The norm today is for delivery drivers to leave packages and take a proof of delivery photo that is then sent to the consumer confirming the successful delivery. The robot already has a camera (it’s eyes) and it’s connected to a 5G network, so the second the robot leaves the package on the doorstep, it snaps a photo and hey presto the proof of delivery jpeg is sent to the consumers phone / email. By the time the customer opens their door to retrieve the package, the robot is long gone and the consumer is none the wiser (unless they check their connected doorbell footage) that the terminator took a visit to their house. Perhaps society is not ready for this, but if it results in lower costs and faster shipping for online orders then I’m sure most consumers will get used to it over time. Maybe in the future, an option in the Shopify checkout will be a $10 “white glove, human delivery” fee.
- Theft and abuse. Teenagers and perhaps other demographic groups may take pleasure in provoking robots, trying to steal the payload, kicking them (for a laugh) or kidnapping them (for ransom or resale). These things will inevitably happen and like a human, these robots will need defensive mechanisms. They will be connected to a 5G network, have 360o camera footage and a GPS tracker. So any attempt to “engage” a robot can be recorded with the footage automatically streamed to the cloud, monitored remotely in the carrier’s control center and when necessary passed onto law enforcement agencies. If the robot is kidnapped it can report its location. I would also argue that the bigger the robot (i.e. a humanoid) the less likely it is to be targeted. The old “pick on someone your own size” saying comes to mind.
- Safety in the public environment. This is the big one. Just like Tesla, Waymo and Cruise have had to prove to state and federal regulars that their self-driving cars are safe to the public, the same will be true of canine or humanoid delivery robots. These robots have very powerful hydraulic actuators that could cause serious injury to a child should for example they get their fingers trapped while interacting with the robot. In the case of an attack on a robot (i.e. an attempted theft of a package), who would be liable if the robot caused an injury or even the death of an assailant? These are all questions the regulators and legal system will need to figure out, but given they have done this for self-driving cars, then I’m pretty confident there is a path forward to regulate delivery robots on our sidewalks. The use case is very similar and I’d argue the overall risk to the public is less for delivery robots walking on the sidewalks vs autonomous vehicles on the road.
- Dogs. This is a difficult one. Perhaps the bigger the robot, the less the dog threat matters. Even a big dog couldn’t do any damage to a full sized humanoid robot and maybe having robots is better anyway, they don’t care if they get bitten, whereas the UPS driver does. The issue will be liability again, if the robot (accidently) hurts the dog, then who is at fault?
Robot’s Can’t Drive The Van – Doh!
Now, some of you still reading this post are probably thinking, this is all very well and good, but someone still has to drive the delivery van, park it outside my house and then drive it onto the next house and so on and so on. Although R2D2 could fly the X-Wing fighter, I’m not sure the wheeled robot can drive a van, nor probably can the canine robot and I don’t either that driving a car is currently part of the Tesla AI team’s roadmap.
So for now let’s assume the delivery van still needs a human driver. If that’s the case, then there is no ROI in any of this. The business case for autonomizing the last mile is dead in the water.
The answer is that we simultaneously need to solve two problems. We need both delivery robots and autonomous vehicles. One is useless without the other.
At the carrier depot (the pickup point), not much changes from a process and operations POV. Autonomous delivery vans would still be manually loaded (for now) with all the packages for the delivery route in just the same way they are today. The difference would be that the delivery van would utilize self-driving technology (presumably licensed from Tesla, Weymo or Cruise) that enables the van to autonomously drive itself to a neighborhood street, park by the sidewalk and pop on its hazard lights.
This is where the canine or humanoid robot (or perhaps multiple robots) spring into action. Inside the van, they pickup and scan the package (no need for a handheld scanner, they can scan with their eyes – very clever), the rear door autonomously opens, the robot jumps out (que a parkour roll for good measure and great social media footage), the door closes again for security and the robot walks down the street to the house / apartment it’s performing the delivery to. If there are multiple homes all with delivery nearby, the robot walks back to the van, retrieves the next package and repeats until it’s time to hitch a lift as the van autonomously drives to the next street.
The great thing here is that the robots can recharge their batteries directly from the van (assuming it’s an electric van like a Rivian, Canoo, BrightDrop or Arrival – which the industry is already moving towards) ensuring they are topped up and able to deliver packages all day long without any battery charge anxiety issues.
The limiting factor on how many deliveries the van and robot can make in a day will not be based on battery life, it will be determined by what society (and cities / states) consider to be socially acceptable delivery hours (currently assumed to be 8am to 9pm) and the volume capacity of the vans themselves. In theory if consumers can accept something resembling the Terminator walking down their drive at 3am, autonomous last mile delivery could occur 24×7, but let’s not get ahead of ourselves.
Maybe automating the last mile just isn’t the answer
Maybe we’re all barking up the wrong tree and chasing shiny objects. Maybe the way to lower last mile delivery costs isn’t to replace human delivery drivers with drones or robots but instead we need to focus on other (short to medium) term solutions that can bring down the cost of last mile deliveries.
So what might they be?
#1 Bionic Assistance
Perhaps the human delivery driver is irreplaceable and what the industry needs to focus on is making them more efficient (more deliveries per hour). In urban areas, much time is wasted finding somewhere to park the delivery van, so increasingly final mile carriers are experimenting with e-assisted wagons, e-bikes and other mobility devices that allow the driver to walk (or ride) a route with more packages than could be carried physically. Two examples of such trials are with UPS – one in London is underway with UPS and Fernhay in London that allows the delivery driver to pull / push an e-assisted cart along their route while another in New York provides an quad wheeled e-bike to the “ridden” by the driver. In both cases the bionic assistance enables the driver to carry far more packages and weight than they could ergonomically carry themselves in a backpack or satchel. Therefore the number of packages they can theoretically deliver before they need to go back to the van / depot and reload is much higher resulting in more deliveries per driver per hour.
#2 Electric Delivery Vans
After labor, the 2nd most significant cost of performing the last mile delivery is fuel. Driving diesel trucks around in a stop – start – stop – start manner in an urban / residential setting is incredibly bad for fuel economy and isn’t too good for the environment either.
Moving their fleets to battery electric vehicles provide both cost savings, environmental and ergonomic advantages to delivery carriers:
- Fuel savings. Moving from diesel to electric can cut the cost of fuel by up to 80% (assuming the vans are charged overnight at the carrier depot or at the drivers home) or by 100% in the case of carriers that invest in their own sustainable electric generation capabilities by putting solar cells on the roofs of their facilities and investing in stationary storage to allow the vans to charge overnight from electricity generated by the sun during the day.
- Ergonomic advantages. With Electric vans, drivers can leave the AC and heating running all the time (no need to leave the engine idling) – this is a huge win for the drivers to have a comfortable work environment even in extreme temperatures
So let’s do a reality check.
Whatever the future of last mile automation is or isn’t, It’s highly unlikely that UPS or the USPS will lead the innovation charge. UPS’s most recently negotiated contract with the teamsters union states that if UPS wants to implement drones, vehicle platooning or driverless pickup and delivery vehicles, the company must notify the Teamsters’ National Negotiating Committee six months in advance of the change. It also “shall be required to bargain the effects of any such change.” Basically UPS has no chance of implementing any autonomous last mile technology that would replace a human driver during the next 5 years. I also find it very unlikely that the government funded USPS would tolerate mass layoffs of government postal worker employees.
The level of investment required to make last mile autonomous deliveries a commercial reality will be staggering and run into billions of dollars just for the R&D, before any capital investment to actually rollout a solution at scale across the US. However the reward for the companies that can do this will be high, so this is exactly the type of investment that VC firms will continue to make bets on over the next decade.
I’m placing my chips on Tesla being a leading supplier of the technology required to make this happen. The question is not if, it’s when. Normally this kind of technology revolution typically takes at least a decade to mature and even when it does, it won’t be a one size fits all answer. But Tesla is not normal. They are without argument the most innovative company in recent US history, they obsess about solving complex engineering challenges, their AI technology is already 10 years mature (inherited from their Automotive Autopilot program) and although Elon has a history of over promising on release dates, he always delivers. Tesla already has many of the pieces of the puzzle that will be required:
In just 2 years, Tesla has taken their Bot from a concept in Elon’s head to a viable, working prototype. I’d argue they they realize pretty quickly that there is bigger (short term) opportunity selling these Bot’s to VC backed delivery startups than there is to end consumers. Between UPS, FedEx and USPS there are over 1.5m delivery drivers in the US today. Assuming each Tesla bot initially has a price of $200k, that alone is a $300 billion addressable market for Tesla.
In addition Tesla already have a mature self-driving technology platform with their Autopilot. There are many rumors that Tesla is already developing a van to add to their vehicle lineup and even if the van doesn’t materialize it is highly likely Tesla start to license their Autopilot technology to other vehicle manufactures. Furthermore the RoboTaxi could play a huge role here, with consumer owned vehicles making themselves available for autonomous parcel delivery when not required.
Drones may very well find a niche delivering to remote, rural addresses that today are only served by the postal service, wheeled robots will likely continue to deliver some pizza’s and groceries on university campuses and autonomous h-vtol aircraft may start to replace trucks for moving parcels between DC’s, but none of these technologies create any future job threat for the UPS delivery driver.
However, I do predict that within 10 years we will start to see humanoid robots on the sidewalks of America’s streets, carrying parcels to our doorsteps and hitching a ride on autonomously operated, electric delivery vans. This may seem far fetched, but no one predicted smartphones or social media 10 years before they became mainstream either.
It’s not a question of if, it’s a question of when. The technology to support a fully autonomous last mile is getting closer, the bigger issue is where the capital required to invest in this change will come from. I don’t think UPS or FedEx shareholders have the foresight to make that bet, but I know a few VC firms that probably will.
In the meantime UPS drivers have nothing to fear. They will continue to earn $170k / year, but the day will come when their job will become obsolete and they will be replaced by a robot. The growth of ecommerce depends on it.