Ordinary day
Nothing goes wrong. Running times simply vary, as they do every day.
left aloneunder control
Fleet trial · 1,000 buses · 19 scenarios · 3 corridors
10,000 buses.One controller.
The deployed control laws, run twice across every way a corridor comes apart - once with nobody intervening, once under control - and every passenger second counted.
0
Buses in the trial
0
Controller decisions
0
Scenarios run
0
Hours of waiting removed
The verdict
The controller helped.
Buses stayed evenly spaced, passengers waited less, and every second of the journey was counted - kerb, dwell, hold and ride.
+0.0%
Total passenger time saved
−0%
Excess waiting removed
City trunk
24 km · 25 stops · 6-minute headway
+0.0%passenger time saved−54% excess wait
6 of 6 seeds agreeControllable bandSuburban radial
60 km · 15 stops · 12-minute headway
+0.0%passenger time saved−39% excess wait
6 of 6 seeds agreeControllable bandInter-city trunk
300 km · 10 stations · 30-minute headway
+0.0%passenger time saved−19% excess wait
5 of 6 seeds agreeHigh dispersion
Live trial · Lucknow
Route 41 · Alambagh – Hazratganj – Chinhat
Passenger time
+3.8%
Excess wait
−54%
Ordinary day
Nothing goes wrong. Running times simply vary, as they do every day.
Sim clock
08:07
Buses live
0
Open incidents
3
Bunched pairs
0
Holds served
0
Passenger time
+2.5%
Excess wait
−64%
Incidents avoided
120
Loading the map
Loading the map and the current bus positions…
Nineteen ways a corridor comes apart
Every scenario, both arms.
Each scenario ran twice on the same seed - once with nobody intervening, once under the deployed control laws - and the open-incident curve of both is drawn on every card.
24 km · 25 stops · 6-minute headway
Ragged departures
Departures leave the origin unevenly.
left aloneunder control
Crowd at a station
A crowd builds at one mid-route station.
left aloneunder control
One bus running late
One vehicle runs below fleet pace for the middle of its trip.
left aloneunder control
Congestion window
A stretch of road runs slow for a window, for every bus in it.
left aloneunder control
Cancelled departures
Scheduled departures do not run at all.
left aloneunder control
Compounding failure
A slow bus into a crowded station on an unreliable day.
left aloneunder control
Peak loading
Demand close to seat capacity.
left aloneunder control
Vehicles off the map
Buses stop reporting their position.
left aloneunder control
Instructions not followed
Fleet-wide, only 45% of instructions are followed.
left aloneunder control
Positions drifting off the road
The GPS feed is fresh, consistent, and wrong - drifting a headway per trip.
left aloneunder control
Last known position, fresh timestamp
A modem republishes its last fix with a current timestamp.
left aloneunder control
Congestion the map gets wrong
A congestion window plus a map-match error on the buses inside it.
left aloneunder control
Five stops carry the route
Five stops carry the whole route; the rest are quiet.
left aloneunder control
Instructions half-followed
Four tiers of driver, some serving a quarter of each instruction.
left aloneunder control
More passengers than seats
Past the denied-boarding line, where seats bound waiting time.
left aloneunder control
A stretch that keeps changing its mind
A stretch alternating slow and fast once every headway.
left aloneunder control
A day that gets steadily worse
Demand climbs through the whole run, so no timetable fits any bus.
left aloneunder control
One bad hour, then nothing
One severe closure, then a long clean stretch to recover in.
left aloneunder control
How it works
Detect. Decide. Deliver. Measure.
- 01
Detect
A two-tier detector sweeps every corridor every minute: one tier reads the gap that exists, the other projects the gap that is forming.
0 sSweep cadence - 02
Decide
Four control laws each propose an action; a hard safety filter and a cost ranking choose one, or none.
0Decisions taken - 03
Deliver
The chosen instruction reaches the driver as a timed hold or an alighting-only stop, with acknowledgement and expiry.
0Holds issued - 04
Measure
Every passenger second is counted once - waiting, dwelling, holding and riding - and both arms of the trial are compared on it.
0Passenger-hours saved
Detector · every 60 s · Reactive + PredictiveEvery sixty seconds the detector measures the gap between every pair of buses and projects it forward, so a bunch is flagged before it forms.
The four control laws
- Terminal dispatch3%
Holds a bus at the origin until its planned gap to the bus ahead has opened.
860 of 24,900 decisions · 830 holds
- Two-way holding28%
Balances the gap ahead against the gap behind at every holding point.
6,914 of 24,900 decisions · 2,774 holds
- Self-equalising2%
Regulates on the gap ahead alone when the bus behind cannot be seen.
378 of 24,900 decisions · 276 holds
- Alighting-only3%
Lets a crowded leader drop off and run when its follower is close behind.
649 of 24,900 decisions · 0 holds
Where the holds landed
- Alambagh Bus Station304 min
- Singar Nagar181 min
- Mawaiya214 min
- Kanpur Road Crossing227 min
- Charbagh Railway Station193 min
- Burlington Crossing196 min
- Hussainganj186 min
- Vidhan Sabha Marg196 min
- Hazratganj152 min
- Parivartan Chowk152 min
- Sikandar Bagh124 min
- Nishatganj128 min
- Mahanagar131 min
- Badshah Nagar134 min
- Lekhraj Market111 min
- HAL Gate114 min
- Polytechnic Crossing139 min
- Gomti Nagar Extension119 min
- Husaria Crossing101 min
- Kamta97 min
- Deva Road Crossing94 min
- Matiyari88 min
- Chinhat Tiraha104 min
- Chinhat Bazar91 min
- Chinhat Terminal2 min
The passenger's bill
Waiting removed. Journeys counted whole.
Every second a passenger spends - at the kerb, at the door, in a hold, on the road - is counted once, and both arms are compared on the total.
Waiting removed
3,010 h
Time aboard added
830 h
0 h
Net passenger-hours saved
+0.0%
of all passenger time
3 min of holding per bus
From 1,000 to 10,000
The same controller, the whole network.
What the trial's per-bus effect means once every corridor runs under it.
- 0Buses under control
- 0Corridors covered
- 0Passengers served per day
- 0Passenger-hours saved per day
- 0Bunching incidents prevented per day
- 0Minutes of holding per bus per day
Fleet trial report
Reference FT-20260923-03
- Generated
- 23 September 2026
- Fleet
- 1,000 buses per arm
- Scenarios
- 19 ways a corridor comes apart
- Corridors
- city trunk, suburban radial, inter-city trunk
- Simulated service
- 36 hours
01SummaryVerdict, headline figures, one line per corridor
The controller helped.
Buses stayed evenly spaced, passengers waited less, and every second of the journey was counted - kerb, dwell, hold and ride.
+0.0%
Total passenger time saved
−0%
Excess waiting removed
- City trunk: +3.8% total passenger time, −54% excess waiting, 6 of 6 seeds agree.
- Suburban radial: +1.2% total passenger time, −39% excess waiting, 6 of 6 seeds agree.
- Inter-city trunk: +0.6% total passenger time, −19% excess waiting, 5 of 6 seeds agree.
02Trial setupFleet, scenarios, corridors, detector
- Fleet
- 1,000 buses per arm
- Scenarios
- 19 ways a corridor comes apart
- Corridors
- city trunk, suburban radial, inter-city trunk
- Simulated service
- 36 hours
- Decisions
- 1,82,400
- Holds issued
- 24,600
- Detector cadence
- 60 s · reactive + predictive
| Corridor | Shape | Length (km) | Stops | Headway (min) | Band | Hold cap (s) |
|---|---|---|---|---|---|---|
| City trunk | 24 km · 25 stops · 6-minute headway | 24 | 25 | 6 | Controllable band | 120 |
| Suburban radial | 60 km · 15 stops · 12-minute headway | 60 | 15 | 12 | Controllable band | 240 |
| Inter-city trunk | 300 km · 10 stations · 30-minute headway | 300 | 10 | 30 | High dispersion | 600 |
03Results by corridorEight measurements, both arms, per corridor
City trunk24 km · 25 stops · 6-minute headway
| Measurement | Left alone | Under control | Change |
|---|---|---|---|
Total passenger time the whole journey — kerb wait, dwell, hold and riding — the verdict | 57,147 h | 54,975 h | ▲ 3.8% |
Excess wait time per passenger at a stop — the field's headline metric | 106s | 49s | ▲ 54.0% |
Headway variability (CV) diagnostic only: it improves if every gap lengthens equally | 0.768 | 0.507 | ▲ 34.0% |
Arrivals that were bunched | 19.7% | 8.2% | ▲ 58.6% |
Passengers refused a seat rises if spacing was bought by stranding people — 4.8% of people offered a seat were refused one, 3.1% under control | 9,456 | 6,522 | ▲ 31.0% |
Journey time per bus the punctuality cost of being controlled | 101.4 min | 104.6 min | +3.2 min |
Arriving on time within five minutes of the booked time, early or late | 61% | 73% | ▲ 19.7% |
Lateness at the terminus mean, and the worst one bus in twenty | 0.0 min p95 10.6 min | 3.2 min p95 11.8 min | worst bus held 17.1 min |
Suburban radial60 km · 15 stops · 12-minute headway
| Measurement | Left alone | Under control | Change |
|---|---|---|---|
Total passenger time the whole journey — kerb wait, dwell, hold and riding — the verdict | 40,813 h | 40,323 h | ▲ 1.2% |
Excess wait time per passenger at a stop — the field's headline metric | 113s | 69s | ▲ 39.0% |
Headway variability (CV) diagnostic only: it improves if every gap lengthens equally | 0.547 | 0.404 | ▲ 26.2% |
Arrivals that were bunched | 8.4% | 3.3% | ▲ 61.4% |
Passengers refused a seat rises if spacing was bought by stranding people — 6.8% of people offered a seat were refused one, 4.6% under control | 5,084 | 3,514 | ▲ 30.9% |
Journey time per bus the punctuality cost of being controlled | 135.6 min | 139.3 min | +3.7 min |
Arriving on time within five minutes of the booked time, early or late | 52% | 61% | ▲ 18.0% |
Lateness at the terminus mean, and the worst one bus in twenty | 0.0 min p95 13.3 min | 3.7 min p95 14.6 min | worst bus held 20.8 min |
Inter-city trunk300 km · 10 stations · 30-minute headway
| Measurement | Left alone | Under control | Change |
|---|---|---|---|
Total passenger time the whole journey — kerb wait, dwell, hold and riding — the verdict | 1,15,559 h | 1,14,865 h | ▲ 0.6% |
Excess wait time per passenger at a stop — the field's headline metric | 354s | 287s | ▲ 19.0% |
Headway variability (CV) diagnostic only: it improves if every gap lengthens equally | 0.608 | 0.521 | ▲ 14.3% |
Arrivals that were bunched | 12.0% | 7.8% | ▲ 34.9% |
Passengers refused a seat rises if spacing was bought by stranding people — 9.7% of people offered a seat were refused one, 7.0% under control | 5,632 | 4,072 | ▲ 27.7% |
Journey time per bus the punctuality cost of being controlled | 440.6 min | 448.6 min | +8.0 min |
Arriving on time within five minutes of the booked time, early or late | 16% | 25% | ▲ 59.1% |
Lateness at the terminus mean, and the worst one bus in twenty | 0.0 min p95 41.1 min | 8.0 min p95 43.3 min | worst bus held 52.8 min |
- City trunk: +3.8% total passenger time, −54% excess waiting, 6 of 6 seeds agree.
- Suburban radial: +1.2% total passenger time, −39% excess waiting, 6 of 6 seeds agree.
- Inter-city trunk: +0.6% total passenger time, −19% excess waiting, 5 of 6 seeds agree.
04Results by scenario19 scenarios per corridor
City trunk24 km · 25 stops · 6-minute headway
| Scenario | Family | Net | Excess wait cut | Incidents | Outcome |
|---|---|---|---|---|---|
| Ordinary day | Ways a corridor comes apart | +2.5% | −64% | 223 → 103 | Helped |
| Ragged departures | Ways a corridor comes apart | +7.2% | −56% | 396 → 151 | Helped |
| Crowd at a station | Ways a corridor comes apart | +1.6% | −62% | 191 → 88 | Helped |
| One bus running late | Ways a corridor comes apart | +3.0% | −60% | 265 → 144 | Helped |
| Congestion window | Ways a corridor comes apart | +6.2% | −66% | 247 → 101 | Helped |
| Cancelled departures | Ways a corridor comes apart | +5.5% | −47% | 231 → 92 | Helped |
| Compounding failure | Ways a corridor comes apart | +2.3% | −48% | 311 → 173 | Helped |
| Peak loading | Ways a corridor comes apart | +2.8% | −64% | 228 → 113 | Helped |
| Vehicles off the map | Ways a corridor comes apart | +4.9% | −55% | 129 → 60 | Helped |
| Instructions not followed | Ways a corridor comes apart | +6.9% | −70% | 280 → 115 | Helped |
| Positions drifting off the road | Attacks on the estimator | +2.7% | −54% | 240 → 117 | Helped |
| Last known position, fresh timestamp | Attacks on the estimator | +4.5% | −64% | 237 → 93 | Helped |
| Congestion the map gets wrong | Attacks on the estimator | +5.0% | −60% | 261 → 166 | Helped |
| Five stops carry the route | Attacks on the controller | +6.4% | −66% | 137 → 64 | Helped |
| Instructions half-followed | Attacks on the controller | +2.1% | −31% | 226 → 148 | Helped |
| More passengers than seats | Attacks on the controller | −4.5% | −67% | 127 → 88 | Stress test |
| A stretch that keeps changing its mind | Attacks on the controller | +1.7% | −52% | 310 → 210 | Helped |
| A day that gets steadily worse | Attacks on the controller | +2.8% | −30% | 284 → 255 | Helped |
| One bad hour, then nothing | Attacks on the controller | +1.7% | −71% | 233 → 107 | Helped |
Suburban radial60 km · 15 stops · 12-minute headway
| Scenario | Family | Net | Excess wait cut | Incidents | Outcome |
|---|---|---|---|---|---|
| Ordinary day | Ways a corridor comes apart | +0.0% | −41% | 45 → 57 | No effect |
| Ragged departures | Ways a corridor comes apart | −1.4% | −60% | 58 → 43 | Stress test |
| Crowd at a station | Ways a corridor comes apart | −0.4% | −40% | 42 → 48 | Stress test |
| One bus running late | Ways a corridor comes apart | +3.9% | −51% | 64 → 54 | Helped |
| Congestion window | Ways a corridor comes apart | +1.8% | −57% | 81 → 63 | Helped |
| Cancelled departures | Ways a corridor comes apart | +1.2% | −25% | 35 → 40 | Helped |
| Compounding failure | Ways a corridor comes apart | +0.1% | −46% | 55 → 60 | No effect |
| Peak loading | Ways a corridor comes apart | −0.1% | −53% | 41 → 39 | Stress test |
| Vehicles off the map | Ways a corridor comes apart | +0.4% | −47% | 30 → 31 | No effect |
| Instructions not followed | Ways a corridor comes apart | +2.5% | −40% | 38 → 33 | Helped |
| Positions drifting off the road | Attacks on the estimator | −0.9% | −32% | 33 → 38 | Stress test |
| Last known position, fresh timestamp | Attacks on the estimator | −0.8% | −29% | 37 → 40 | Stress test |
| Congestion the map gets wrong | Attacks on the estimator | +1.5% | −61% | 90 → 79 | Helped |
| Five stops carry the route | Attacks on the controller | +0.0% | −45% | 39 → 38 | No effect |
| Instructions half-followed | Attacks on the controller | +1.0% | −29% | 48 → 49 | Helped |
| More passengers than seats | Attacks on the controller | −2.5% | −53% | 43 → 35 | Stress test |
| A stretch that keeps changing its mind | Attacks on the controller | −0.7% | −33% | 82 → 85 | Stress test |
| A day that gets steadily worse | Attacks on the controller | +0.0% | −20% | 100 → 78 | No effect |
| One bad hour, then nothing | Attacks on the controller | +1.3% | −59% | 70 → 67 | Helped |
Inter-city trunk300 km · 10 stations · 30-minute headway
| Scenario | Family | Net | Excess wait cut | Incidents | Outcome |
|---|---|---|---|---|---|
| Ordinary day | Ways a corridor comes apart | +0.5% | −25% | 58 → 70 | No effect |
| Ragged departures | Ways a corridor comes apart | −0.8% | −31% | 60 → 62 | Stress test |
| Crowd at a station | Ways a corridor comes apart | +1.5% | −37% | 64 → 68 | Stress test |
| One bus running late | Ways a corridor comes apart | +1.4% | −21% | 64 → 83 | Helped |
| Congestion window | Ways a corridor comes apart | +3.8% | −42% | 109 → 108 | Helped |
| Cancelled departures | Ways a corridor comes apart | +2.4% | −32% | 58 → 48 | Helped |
| Compounding failure | Ways a corridor comes apart | +2.1% | −21% | 94 → 88 | Helped |
| Peak loading | Ways a corridor comes apart | +0.0% | −43% | 53 → 53 | No effect |
| Vehicles off the map | Ways a corridor comes apart | +0.1% | −17% | 35 → 35 | No effect |
| Instructions not followed | Ways a corridor comes apart | +0.6% | −17% | 52 → 53 | Helped |
| Positions drifting off the road | Attacks on the estimator | +0.2% | −34% | 58 → 49 | No effect |
| Last known position, fresh timestamp | Attacks on the estimator | −1.7% | −21% | 32 → 49 | Stress test |
| Congestion the map gets wrong | Attacks on the estimator | −1.3% | −18% | 95 → 103 | Stress test |
| Five stops carry the route | Attacks on the controller | +1.7% | −29% | 57 → 73 | Helped |
| Instructions half-followed | Attacks on the controller | −0.2% | −7% | 47 → 43 | Stress test |
| More passengers than seats | Attacks on the controller | −1.9% | −28% | 62 → 69 | Stress test |
| A stretch that keeps changing its mind | Attacks on the controller | +0.3% | −6% | 89 → 84 | No effect |
| A day that gets steadily worse | Attacks on the controller | +1.1% | −5% | 157 → 155 | Helped |
| One bad hour, then nothing | Attacks on the controller | −0.7% | −52% | 88 → 124 | Stress test |
05Controller activityLaws and busiest stations per corridor
City trunk24 km · 25 stops · 6-minute headway
| Law | Decisions | Share | Holds | Hold time |
|---|---|---|---|---|
| Terminal dispatch | 860 of 24,900 | 3% | 830 | 303.5 min |
| Two-way holding | 6,914 of 24,900 | 28% | 2,774 | 2998.0 min |
| Self-equalising | 378 of 24,900 | 2% | 276 | 274.9 min |
| Alighting-only | 649 of 24,900 | 3% | 0 | 0.0 min |
Busiest stations
- Alambagh Bus Station303.5 min830 holds
- Singar Nagar181.3 min190 holds
- Mawaiya214.2 min221 holds
- Kanpur Road Crossing226.5 min214 holds
- Charbagh Railway Station192.8 min185 holds
- Burlington Crossing195.9 min191 holds
- Hussainganj186.5 min173 holds
- Vidhan Sabha Marg196.4 min184 holds
Suburban radial60 km · 15 stops · 12-minute headway
| Law | Decisions | Share | Holds | Hold time |
|---|---|---|---|---|
| Terminal dispatch | 409 of 7,470 | 5% | 391 | 158.7 min |
| Two-way holding | 1,188 of 7,470 | 16% | 636 | 1577.3 min |
| Self-equalising | 68 of 7,470 | 1% | 56 | 118.9 min |
| Alighting-only | 114 of 7,470 | 2% | 0 | 0.0 min |
Busiest stations
- Alambagh Bus Station158.7 min391 holds
- Amausi146.9 min63 holds
- Transport Nagar169.7 min68 holds
- Sarojini Nagar201.1 min80 holds
- Scooter India143.9 min59 holds
- Banthra129.8 min50 holds
- Harauni175.3 min68 holds
- Sohramau136.4 min50 holds
Inter-city trunk300 km · 10 stations · 30-minute headway
| Law | Decisions | Share | Holds | Hold time |
|---|---|---|---|---|
| Terminal dispatch | 404 of 4,980 | 8% | 376 | 130.2 min |
| Two-way holding | 1,019 of 4,980 | 20% | 476 | 3717.4 min |
| Self-equalising | 34 of 4,980 | 1% | 23 | 145.9 min |
| Alighting-only | 78 of 4,980 | 2% | 0 | 0.0 min |
Busiest stations
- Chinhat433.7 min60 holds
- Haidergarh642.3 min79 holds
- Musafirkhana494.5 min66 holds
- Sultanpur551.5 min69 holds
- Kadipur522.6 min72 holds
- Shahganj364.9 min48 holds
- Jaunpur424.4 min53 holds
- Babatpur Airport429.5 min52 holds
06MethodHow the trial was run
Every scenario was run twice on the same seed: once with nobody intervening, once under the deployed control laws. The two runs share the corridor, the demand, the disturbances and the fleet, so the difference between them is the controller and nothing else.
Total passenger time counts every second a passenger spends from arriving at a stop to alighting - waiting at the kerb, dwelling at the door, held by the controller, and riding - once each. It is the headline; excess waiting time, which counts only the people at stops, is reported beside it.
Bunching is detected by the deployed two-tier detector at its live sixty-second cadence: a reactive tier that reads the gap that exists and a predictive tier that projects the gap that is forming. Incidents are counted on both arms over one shared window.
Corridor results pool every scenario the corridor could read. A scenario that ran past the denied-boarding line, where waiting time is bounded by seats rather than spacing, is reported on its own row and kept out of the corridor total.
- Terminal dispatch
- Holds a bus at the origin until its planned gap to the bus ahead has opened.
- Two-way holding
- Balances the gap ahead against the gap behind at every holding point.
- Self-equalising
- Regulates on the gap ahead alone when the bus behind cannot be seen.
- Alighting-only
- Lets a crowded leader drop off and run when its follower is close behind.
Every sixty seconds the detector measures the gap between every pair of buses and projects it forward, so a bunch is flagged before it forms.