ROBOTAXI, POTHOLES AND COST.

 








GROK:


I’ll look up how Tesla’s current vision-only FSD and robotaxi stack actually handle road-surface hazards like San Francisco potholes, then lay out the logical steps the system would take.

Here’s the logical sequence a vision-only FSD / robotaxi would follow to dodge potholes in a city like San Francisco — and where that sequence still breaks today.

As of Sept 1, 2026, dedicated pothole avoidance is still listed as an “upcoming” item in FSD v14.3 notes. Musk said “coming soon” yesterday. Some builds already occasionally straddle a big hole. It is not a reliable, shipped behavior.

1. See the surface, not just objects

Cameras feed an occupancy network plus a surface model. That model is supposed to estimate:

  • elevation / depth of the road
  • whether a patch is navigable
  • features: curbs, speed bumps, manholes, potholes, patched asphalt

Potholes are hard because they are negative space — a hole in a surface the car already thinks is “road.” 

Shadows, wet pavement, and SF’s patched black-on-black asphalt look a lot like holes. 

Hills hide what’s over the crest until late.

2. Score severity in real time

Not every dip is worth dodging. The system has to estimate:

  • width vs track width (can it pass between the tires?)
  • depth / sharpness (comfort vs wheel damage vs “this might rip a control arm”)
  • distance and time-to-impact
  • lighting confidence (night, glare, rain)

A shallow crack gets ignored. 

A tire-eater gets a high cost.

3. Check whether the planned path hits it

The planner already has a trajectory. 

It asks: 

do my left tires, right tires, or both intersect this voxel?

If yes, generate options:

  • Straddle — small lateral nudge so the hole passes under the belly (this is the behavior already seen in some v14 clips)
  • Slow — reduce speed so the hit is a thump, not a slam
  • Lane-offset — stay in lane but hug the cleaner side
  • Lane change — only if adjacent lane is clear and legal
  • Stop / creep — last resort if the whole lane is destroyed and traffic allows

Robotaxi has no steering wheel and no pedals, so there is no “human backup.” 

The planner has to pick one of those and live with it.

4. Apply city constraints (this is the SF part)

San Francisco adds extra filters:

  • Don’t swerve into a bike lane or a parked car
  • Don’t cross a double-yellow unless the road is empty
  • Don’t surprise a cable-car track or a steep crest
  • Don’t make a sharp dodge at 25 mph next to pedestrians
  • Prefer slowing over a dramatic weave on a narrow Mission/SOMA street

A good planner treats “don’t hit the pothole” as a soft cost, not a hard rule. 

Hitting a 2-inch hole is better than clipping a cyclist.

5. Execute smoothly, then remember it

If the car has air/adaptive suspension (Cybertruck; older S/X), it can also raise or soften before impact. 

Most robotaxis won’t have that, so path + speed is the whole toolkit.

After the encounter, the fleet can drop a pin on a shared “rough road” map — Tesla already does a version of this for suspension. 

Next cars get earlier warning and can change lanes before the hole is in camera range. 

That matters more in SF than live vision alone, because hills and parked vans block the view.

What actually happens today

  • Large, obvious holes: sometimes straddled
  • Small / shadowed / rain-filled holes: often driven through
  • No dedicated, consistent “avoid potholes” policy in current supervised FSD
  • Unsupervised Cybercab still has to solve this without a human grabbing the wheel

So the logical stack is: 

see surface → score hole → choose straddle / slow / shift → respect traffic → write it to the fleet map. 

The missing piece is making step 1 reliable enough that step 3 isn’t a coin flip on a shadowed SF street. 


That’s why the feature has been “coming soon” since 2019 and is still in the release notes.


I’ll pull typical tire/rim repair costs and how often robotaxis would hit that damage, then turn it into a per-mile and per-vehicle cost factor.

Tire and rim damage is one of the few opex lines that can quietly eat a robotaxi’s margin. 

Energy on a Cybercab is only about 2.6¢/mile. 

Tires plus pothole hits can land in the same neighborhood if the car does not dodge holes.

Unit costs (2026, Tesla-class)

ItemTypical costNotes
One Tesla tire$250–$350Model Y range; Cybercab should be cheaper (lighter, FWD, smaller tire)
Full set of 4, installed$1,200–$1,800Consumer Tesla pricing
OEM Tesla wheel$900–$1,200Service-center list
Wheel + tire + labor (one corner)$1,300–$1,600The real “I ate a pothole” bill
Rim straighten/refinish$75–$250Only if not cracked
Alignment after a hit$100–$200Almost always needed
AAA average pothole repair~$400Mix of tire / wheel / alignment

Tesla’s own Wheel and Tire Protection Plan prices that risk at $16–$24/month for consumer cars, with a $25 deductible per replacement visit. 

That is Tesla’s implied insurance price for road-hazard damage.

Two different cost stacks

1. Normal wear (even if FSD never hits a hole)


A robotaxi does 60,000–100,000 miles a year. EV tires already wear ~20–30% faster than ICE tires. City stop-start and extra weight make it worse.

  • Tire life on a high-mileage EV: 25,000–40,000 miles per set
  • At $1,400/set and 30,000 miles: ~4.7¢/mile
  • At $1,000/set (fleet Cybercab tires) and 40,000 miles: ~2.5¢/mile

That wear cost exists whether potholes exist or not.

2. Incident damage (potholes, curbs, debris)


This is the extra factor FSD is supposed to kill.

Reasonable city-fleet assumptions:

  • Soft city (Austin, Phoenix): 1 damaging hit every 15,000–25,000 miles
  • Hard city (San Francisco): 1 damaging hit every 8,000–15,000 miles if the car does not systematically avoid holes
  • Mix of outcomes: 50% tire-only ($300), 30% tire+alignment ($450), 20% bent/cracked wheel (~$1,400)

Blended cost per hit ≈ $550–$650.

That adds:

  • Soft city: 2–4¢/mile
  • SF-like city, no avoidance: 4–8¢/mile
  • SF-like city, good avoidance (straddle / slow / map): 1–2¢/mile

Combined cost factor

ScenarioTires (wear)Pothole/rim hitsTotal tire+rim
Consumer Tesla, careful driver3–4¢0.5–1¢~4¢/mi
Robotaxi, good FSD avoidance2.5–4¢1–2¢~4–6¢/mi
Robotaxi, SF, weak avoidance3–5¢4–8¢~8–12¢/mi
Tesla all-in opex target——20¢/mi total

So tires + rims are roughly 20–30% of Tesla’s $0.20/mile target if FSD works, and can become half the operating cost in a pothole city if it doesn’t. 

Energy is only 2.6¢. Tires can cost more than electricity.

Why robotaxis get hit harder than your Model Y

  • Utilization: 3–5× the annual miles, so the same pothole map is sampled every day.
  • No human dodge: a passenger in a wheel-less Cybercab cannot nudge around a hole the net missed.
  • Downtime: a $300 tire is cheap; a 2-hour service pull on a car that should be earning $1–$1.40/mile is not. Lost revenue on one incident is often $50–$150 on top of parts.
  • Alignment creep: repeated small hits eat tires even when no single rim cracks.

Practical takeaway

Treat 5¢/mile as the base tire-and-rim budget for a well-run Cybercab fleet with working pothole avoidance.


Treat 8–10¢/mile as the SF penalty if vision-only FSD still drives through holes.

That 3–5¢ gap is exactly why “coming soon” pothole avoidance is an opex feature, not a comfort feature. 

On a 80,000-mile year, that gap is $2,400–$4,000 per car — enough to matter at Tesla’s thin robotaxi margin.

Comments

Popular posts from this blog

TESLA MODEL 2 vs ROBOTAXI

SENSOR FUSION ORTHODOXY.

THE CAVEAT FACTOR.