Understanding how humans navigate the last mile & how technology can benefit from it
Its been a while I wanted to sit down and think thru the very much popular “last mile problem”. How it looks like; for navigation, delivery robots and such. This kind of exercise brings me the clarity, helps understand the depth of the opportunity. Its a great exercise to validate the problem space and its impact and potential.
As a refresher, the last mile problem is something that has to do with navigating spaces and environments in the very last mile before the arrival or departure- we can consider mostly outdoors. Consider, case of vehicles and specially programable vehicles ( autonomous, drive assisted, robot fleet) going at 60 Miles Per Hour, it’s 1 min of time to travel the last one mile before arrival. In the context of time, a vehicle takes 4 min and at 7 mph, when just getting ready to make a deceleration & before stop it can take upto 8min of time to plan and execute.
So realistically, the slowest vehicle can take from 4 min to upto 8 min to decide whatever it wants to do in the last mile of stopping at the stop, finding a safe place (for the vehicles and others) to stop or get parked. That 4 min is a lot of time for the computer to perform millions of computations, make complex decisions and arrive at a multi modal decision and inference that will then guide the vehicle’s current position and navigation — this seems like a constrained problem space and a task to solve.
Depth
So what is a last mile problem space made of robot vs human?
When visiting a new place, you have very basic prior information — INFORMATION GAP
Sounds, weather, traffic, occupancy, time of day etc. gives situational context — CONTEXT GAP
When dwelling in a place, layovers needs to be safe — SAFETY GAP
All of the above happens real time — TIME/URGENCY GAP
Cascade of decisions taken with certain accuracy and confidence — RELIABILITY GAP
Decisions to be communicated to the actors and respondents — COMMUNICATION SYSTEM GAP
Above all consistently required for variety of situations — REPEATABILITY AND VARIETY
Opportunity
A human’s Last Mile System:
Human: The preparation, intution, ancipation, prediction + Delegating navigation
How do we correctly and safely navigate to some place where we had never been before? We inherently have a solver that solves for all of the above gaps — INFORMATION, CONTEXT, SAFETY, TIMING, RELIABILITY, COMMUNICATION, REPEATABILIY
What is cool about this is, when we are younger, we are learning this gradually & unknowingly : In the school setting, while getting dropped off around safer corridors, slowly our model auto trains and learns gradually in a few years — its a long time to trust a toddler to grow up in a young man who can be trusted for their ability to go to new places and safely come back. I notice, now that I am grown up, I am fully planning and course correcting — carry the directions plus call a friend for gate access, entyway/parking instructions etc. I make extra effort to make sure I have a safe plan — check on traffic, weather, details to get back home(plan A, Plan B) and so on, I bet things were different, I had missed the trains, forgot to carry return route plan ( pre Google Maps) etc. Event then, what struck me is I always spend more time in prepartions and then leaving navigation task to the software, paper map or a human local expert. Navigation always seemed like a task that is well contained. The preparation, intution, ancipation, prediction that is still complicated. Getting out of tricky situations in unknown places is the most nervous part of it… the last mile that needs to be solved today is not a navigation problem…The last mile problem to me is solving for planning, prediction, anticipation and a lot of course correction.
Human: Generalizing the inteligence, expanding an area of brain.
Now that I have arrived somwehere, depending on what I am going to do in the dwelling area (my friend’s place, the coffee shop, the gym, etc.) I would map out where I should get down of my ride, which gate, stairs or corridors are safe for heavy items (in case I am carrying any). It matters what type of heel I am wearing, how I am feeling and so on. If I had an occupational equipment like stroller :-D and a baby with me then everything goes a notch higher on safety aspect, goes down in the efficiency aspect and so on.
More complex tasks we practice on, savvier we get in terms of dealing with simpler situations. If one can take a peek at people’s brains in the busiest metros of the world, my guess is that we will discover some area of their brain in those people is larger than most of us. And if that is the case, we can hope that by bootstrapping the training for computers in the complex physical and highly social environments, one can get better results.
Human: Revisiting the choices and making adjustments, educating others.
“Experience is a teacher” plays out here very well. We are continuously noting the step by step tasks we have to do in a mechanical work e.g. grab a bottle, if we drop, we label it as failed task, we revisit, correct our behaviour and save it as a training experience. If the implications are harsh, we set goals to improve the model. This happens with anything w.r.t. physical efforts. Yes there are emotions that play a part too…even then our learnings are purely subjective, human rating is very subjective but can they be parameterized and can be defined so that machine can model it? in this case — along with the weights we label our mistakes with.
Human not only learns themselves by their mistakes but also educate others… just to expand on this last part, if a machine can teach another machine in an safe and explainable way that could be another breakthrough we are waiting for !
Impact
The way we see and protect ourselves while preserving the safety of the world around us…tremendously focusing on the task in hand moments before approaching the milestone, refining the outcome, judging the outcome, reviewing the results, celebrating the results…this is very impactful if we can transfer this approach to the machine or program assisting us…so that we feel confident about not just the path we are on in the technology adoption but also in the design choices we are putting front and forward.
Potential
Huge potential for innovation, time saving, fuel saving, traffic management, reduction in hazards, planning the places and corridors, feedback management from such systems to improve the town, city, campuses. Make strides in the peace and ease for the world around us.Make safer corridors for humans, animals, make good use of land and assets…the potential applications are many…
So long… keep imagining! :D
