Promising in theory, fragile under actuated control.
Eco-Approach and Departure broadcasts SPaT and MAP messages so connected vehicles can adjust speed to pass the next signal on green — saving fuel and reducing stops. In practice, signal-phase uncertainty under actuated control collapses the operational envelope, and well-coordinated corridors don't benefit at all.
Two coordinated arterial corridors in Pasadena — Hill Ave (6 intersections) and California Blvd (14 intersections) — drawn from the calibrated I-210 corridor network. RSUs broadcast real-time SPaT and MAP messages. The implemented EAD strategy is deliberately conservative because actuated signal phases have uncertain end times.
SPaT at 10 Hz includes the TimeChangeDetails attribute — startTime, minEndTime, maxEndTime. Under actuated control these are estimates: a phase can Gap-Out early or Max-Out late.
If distance to the leader or the stop bar is under 50 ft, the EAD strategy doesn't activate. Vehicle position updates every 0.1 s; speed averaged across a 2 s rolling window.
If red: only accelerate if remaining red is shorter than free-flow travel time to the stop bar — and accelerate to a higher (but sub-posted) limit if it's shorter than current-speed travel time. If green: accelerate to posted limit if remaining green exceeds free-flow travel time.
Speed update interval is 0.5 s. The vehicle holds the new target until conditions change. By design, the strategy is unwilling to depend on the uncertain phase-end times in the SPaT message.
Pasadena, between Walnut St and Del Mar Blvd. Speed limit 30 mph (35 mph at Del Mar). Dedicated left-turn pockets, protected vs permitted phases, lead-lead vs lead-lag patterns. 3 RSUs at @Walnut, @Colorado, @Del Mar.
Pasadena, between Hudson Ave and Michillinda Ave. Speed limits 30/35 mph on California Blvd, 25 mph on side streets. 5 RSUs at @Lake, @Hill, @Sierra Madre, @San Gabriel, @Rosemead.
EAD's mobility benefit appears only during low-demand off-peak periods. During peak hours, and especially on well-coordinated corridors, it actively makes things worse.
The implemented EAD provides mobility improvement during low-demand windows — for example, 6–8 AM on Saturday on Hill Ave. Outside that band, results trend negative.
During high-demand periods, EAD increases average vehicle delay. The algorithm's platoon-forming behavior conflicts with the corridor's existing coordination plan.
When the baseline is already optimized (e.g., California Blvd weekday 11:00–13:00), EAD's interference produces the most severe degradation. Well-coordinated networks don't need EAD's help.
During the narrow window where EAD helps, it helps most on the less-coordinated stretches — Hill Ave Saturday mornings being the canonical example.
Phase end times under actuated control are estimates with Gap-Out and Max-Out behaviors. EAD algorithms that assume reliable signal predictions cannot operate safely in this regime.
EAD encourages CVs to form platoons that arrive together at intersections. This vehicle-level behavior is independent of the network-level signal coordination plan, and the two can fight each other.
EAD isn't a flawed concept — it's a concept built for a control regime that California's signalized intersections largely don't use. Make actuated signals more predictable, or change the algorithm to embrace uncertainty.
The implemented EAD strategy was deliberately conservative — refusing to act on uncertain TimeChangeDetails values. Even that conservative version interferes with platooning behavior already produced by the corridor's existing coordination plan. A more aggressive EAD algorithm built on GLOSA-style assumptions would be worse: it would act on phase-end predictions that simply aren't reliable under actuated control.
The deeper issue is architectural. EAD is a vehicle-level controller layered on top of a network-level signal controller, and the two have no way to coordinate. Each is locally optimal, jointly suboptimal. Until the two layers talk — via collaborative-control strategies or via SPaT messages whose phase-end times become certainties — EAD will fight against rather than enhance corridor coordination.
Two paths forward emerged from this study. (1) Develop new signal control strategies that make SPaT information reliable — adaptive fixed-time control is the most direct route. (2) Develop collaborative-control strategies where the signal controller and the vehicle controller synchronize their decisions, rather than each optimizing independently.
Do not deploy Eco-Approach & Departure on arterial corridors under coordinated actuated control. The implementation tested either provides no benefit or actively increases delay during peak hours.
Avoid GLOSA and other EAD algorithms in the literature that assume 100% accurate signal-phase information. Their safety properties don't hold under actuated control.
Pursue adaptive fixed-time signal control research as a prerequisite. If SPaT TimeChangeDetails values become reliable, the operational envelope of safe EAD expands substantially.
Develop collaborative-control strategies that synchronize signal-controller decisions with vehicle-controller decisions. Independent optimization at each layer produces the conflicts seen in this study.
If pursuing EAD pilots, restrict initial deployment to fixed-time-controlled corridors or genuinely low-demand off-peak windows where the algorithm's benefits hold and its harms are minimal.