Jam-Absorption Driving (JAD)

• JAD is a traffic-flow control strategy that uses a deliberate slow-in fast-out maneuver upstream of congestion to cut off inflow and dissolve jams.
• It embeds control within a single vehicle’s trajectory, relying on precise prediction and state estimation to balance travel time, fuel consumption, and safety.
• Advanced implementations extend JAD with multi-vehicle support, data assimilation, and haptic shared control to manage both primary jam removal and secondary wave suppression.

Jam-absorption driving (JAD) is a traffic-flow control strategy in which a single vehicle performs a deliberate “slow-in, fast-out” maneuver upstream of a wide moving jam or stop-and-go wave. In its canonical form, the vehicle first slows down and maintains a low velocity; because this cuts off the supply of vehicles to the jam, the jam shrinks and finally disappears. The vehicle then returns to following the vehicle ahead of it. In the stop-and-go-wave suppression literature, JAD is distinguished by its vehicle-embedded character: rather than imposing control through roadside infrastructure, it uses one absorbing vehicle, or a very small controlled subset, to alter macroscopic flow evolution (Nishi, 2019, He et al., 15 Apr 2025).

1. Core operating principle and problem setting

JAD is formulated for upstream-moving congestion structures described variously as wide moving jams, shock waves, or stop-and-go waves. The essential mechanism is inflow regulation at the upstream front of the congestion region: the absorbing vehicle slows upstream of the jam, creates a controllable gap, prevents additional vehicles from feeding the jam, and later accelerates back to normal following once the target disturbance has been absorbed. The literature repeatedly describes this as a “slow-in, fast-out” maneuver, and uses it both for isolated wide moving jams on homogeneous roads and for jams fixed at bottlenecks such as sags (Nishi, 2019, Nishi et al., 2021).

The review literature places JAD alongside variable speed limit (VSL) control as one of the two prominent strategies for stop-and-go wave suppression. The distinction is methodological as well as operational. VSL acts externally through infrastructure, whereas JAD acts internally through a vehicle trajectory embedded in the stream, with continuous spatial and temporal resolution rather than segment-based speed commands. This makes JAD attractive in settings where only a single controlled vehicle is available, but it also makes performance sensitive to prediction accuracy, local instability, and maneuver execution (He et al., 15 Apr 2025).

A persistent misconception is that eliminating the primary jam is equivalent to stabilizing the surrounding traffic. JAD research does not support that simplification. The primary wave may disappear while new perturbations generated by the absorbing maneuver amplify upstream into secondary jams or secondary shock waves. Much of the technical literature on JAD is therefore organized around a dual objective: remove the target jam and restrict secondary-wave formation (Nishi, 2019, Suzuki et al., 5 Aug 2025).

2. String stability and the restriction of secondary jams

The main theoretical treatment of secondary-jam suppression models a semi-infinite, non-periodic, single-lane system consisting of one vehicle performing JAD and upstream human-driven vehicles obeying a car-following model. The key analytical step is to apply linear string stability to the macroscopic spatiotemporal structure induced by JAD. In the Intelligent Driver Model (IDM), the analysis yields a critical velocity $v_{\text{cr}}$ above which the equilibrium flow is linearly string stable. The central JAD restriction is then expressed as

$$v_{\text{a,mac}}=\frac{v_{\text{R}}}{v_{\text{S}}}v_{\text{ini}} \ge v_{\text{cr}},$$

where $v_{\text{ini}}$ is the initial pre-jam velocity, $v_{\text{R}}$ is the velocity of the downstream head of the jam, and $v_{\text{S}}$ is the velocity of the upstream tail. Operationally, the condition says that the absorbing vehicle must not slow so much during the slow-in phase that the upstream platoon enters a string-unstable regime (Nishi, 2019).

The same study reports finite simulations with up to $10^5$ vehicles and classifies outcomes into three regimes: F, where initial perturbations do not lead to wide moving jams; NSJ, where wide moving jams occur but are eliminated by JAD without triggering secondary jams; and SJ, where JAD triggers secondary jams. The NSJ regime is substantial for typical IDM parameter settings, especially when absorbing velocities are not too low. As the initial velocity $v_{\text{ini}}$ increases, the system transitions from SJ to NSJ and then to F. The NSJ region expands with higher acceleration $a$ and reduces with higher comfortable deceleration $b$ or time gap $T$. The reported boundary between occurrence and suppression of secondary jams is accurately predicted by the theoretical condition in large but finite simulations (Nishi, 2019).

The same string-stability logic is extended to more complex semi-infinite systems. With inflows from other lanes, the vacant space created by JAD may be filled by lane-changing vehicles, which reduces the effective absorbing velocity. With bottlenecks, the downstream head is stationary and the macroscopic expression for the absorbing velocity changes accordingly. In both cases, the governing principle remains the same: the effective absorbing velocity should not fall below $$v_{\text{a,mac}}=\frac{v_{\text{R}}}{v_{\text{S}}}v_{\text{ini}} \ge v_{\text{cr}},$$0. This establishes JAD not merely as an ad hoc maneuver, but as a control action whose admissible range can be stated in closed form under explicit stability assumptions (Nishi, 2019).

3. Bottleneck-induced congestion, sag applications, and scale effects

A separate strand of work studies JAD in bottleneck-driven congestion, especially at sags, where a downhill changes into an uphill. In this setting, the jam is fixed at the bottleneck, and JAD is activated repeatedly as new jams form. The modeled system is single-lane and non-periodic, with all vehicles connected. Microscopic motion is described by IDM+, the road-gradient effect is modeled explicitly, and fuel consumption is computed using the EMIT model. For each emerging jam, a vehicle upstream of the jam front is assigned as the absorbing vehicle, a goal position and goal time are predicted, and the vehicle follows a planned slow-in trajectory toward that goal before returning to normal car-following (Nishi et al., 2021).

The principal question in this literature is not whether one intervention can work, but how the strategy behaves as the system size increases. Simulations varying the number of vehicles from 500 to 10,000 show that JAD can reduce the average total travel time per vehicle, with a slightly increasing rate of reduction as system size grows. It can also reduce the average total fuel consumption per vehicle, but in that case the rate of reduction decreases and becomes roughly constant as the system grows. For $$v_{\text{a,mac}}=\frac{v_{\text{R}}}{v_{\text{S}}}v_{\text{ini}} \ge v_{\text{cr}},$$1, the reported maximum reductions are about 97 seconds per vehicle for average total travel time and about 730 grams per vehicle for average total fuel consumption (Nishi et al., 2021).

An important negative result is that minimizing travel time and minimizing fuel consumption are not simultaneously achievable with the same parameter setting. The optimal spatiotemporal scale for travel-time reduction and the optimal scale for fuel reduction become roughly constant, respectively, but they differ from each other. In the reported interpretation, smaller-scale, more frequent JAD activations can minimize travel time even without always completely removing the jam, whereas larger-scale interventions are preferable for fuel economy. This trade-off is central to practice-oriented JAD design because it shifts the question from “does JAD work?” to “which objective is being optimized?” (Nishi et al., 2021).

4. State estimation, absorbing-end-point prediction, and data assimilation

Practical JAD requires more than a theoretical admissible velocity. It also requires reliable prediction of the absorbing end point, namely the space-time point at which the absorbing vehicle should resume normal driving. A recent data-assimilation framework addresses this problem for sag traffic congestion by combining the extended Kalman filter with the cell transmission model, yielding an EKF-CTM architecture for JAD support. The roadway is divided into $$v_{\text{a,mac}}=\frac{v_{\text{R}}}{v_{\text{S}}}v_{\text{ini}} \ge v_{\text{cr}},$$2 cells, loop detector data are collected at cell boundaries, and the filter jointly estimates cellwise density and key fundamental-diagram parameters through the augmented state

$$v_{\text{a,mac}}=\frac{v_{\text{R}}}{v_{\text{S}}}v_{\text{ini}} \ge v_{\text{cr}},$$3

with

$$v_{\text{a,mac}}=\frac{v_{\text{R}}}{v_{\text{S}}}v_{\text{ini}} \ge v_{\text{cr}},$$4

This makes the JAD controller adaptive not only to state uncertainty but also to parameter drift in free-flow speed and critical density (Li et al., 2024).

The motivation is explicit: if the controller relies on misestimated traffic parameters, two failure modes can occur. In underestimation, the absorbing end point is reached too early and the jam is not cleared. In overestimation, the control lasts too long and new jams may be created. In the reported framework, the absorbing end point is recalibrated in real time through state and parameter estimation, and the absorbing vehicle is guided toward that updated target by a control law subject to safe and comfort-based acceleration bounds. Numerical results show that the data-assimilation framework effectively mitigates underestimated or overestimated control failures caused by misestimation of key parameters of the traffic-flow fundamental diagram (Li et al., 2024).

The performance indicators are stated as changes in average travel time, $$v_{\text{a,mac}}=\frac{v_{\text{R}}}{v_{\text{S}}}v_{\text{ini}} \ge v_{\text{cr}},$$5, and fuel consumption, $$v_{\text{a,mac}}=\frac{v_{\text{R}}}{v_{\text{S}}}v_{\text{ini}} \ge v_{\text{cr}},$$6, comparing JAD with and without data assimilation. The reported findings are that travel-time and fuel savings are restored or improved when data assimilation is used, even under initially misestimated model parameters. The framework is also described as robust to changing traffic characteristics such as weather conditions or traffic composition, and as deployable using standard loop detector data rather than dedicated connected-vehicle sensing. This suggests that, in current deployments, state estimation and prediction may be at least as consequential as the nominal control law itself (Li et al., 2024).

5. Multi-vehicle stabilization and human-in-the-loop realizations

One major extension of JAD augments the absorbing vehicle with connected and automated support vehicles (SVs) placed upstream. The rationale is that the absorbing vehicle’s deceleration creates a high-density upstream region in which perturbations can amplify into secondary shock waves. The proposed support-driving (SD) mechanism dynamically extends the IDM time-gap parameter $$v_{\text{a,mac}}=\frac{v_{\text{R}}}{v_{\text{S}}}v_{\text{ini}} \ge v_{\text{cr}},$$7 of selected SVs, with string stability certified through a head-to-tail transfer-function criterion,

$$v_{\text{a,mac}}=\frac{v_{\text{R}}}{v_{\text{S}}}v_{\text{ini}} \ge v_{\text{cr}},$$8

In the reported setup, the stream contains 2000 vehicles, including one absorbing vehicle and five SVs, and a typical large time gap of $$v_{\text{a,mac}}=\frac{v_{\text{R}}}{v_{\text{S}}}v_{\text{ini}} \ge v_{\text{cr}},$$9 is used to achieve strong string stability (Suzuki et al., 5 Aug 2025).

The resulting trade-offs are quantified explicitly. Relative to JAD-only control, the combination of JAD and SD reduces total fuel consumption by 109 kg and collision risk, measured through inverse time-to-collision, by 1940, but increases total travel time by 49.3 hours. If the extended time gap is reverted to its initial value after the jam is cleared, the travel-time penalty is reduced by 27.1 hours while low collision risk is maintained, albeit with increased total fuel consumption by 113 kg relative to the non-reverting support case. A further clarification is important: SD alone cannot eliminate the target shock wave. The primary absorption still requires JAD; the support vehicles stabilize the recovery process and suppress secondary shock waves (Suzuki et al., 5 Aug 2025).

A different realization keeps a human driver in the loop through haptic shared control rather than replacing the driver with full automation. In a driving-simulator experiment with 24 participants on a ring road with 21 cars, a controller based on JAD principles was compared across manual control, haptic shared control, and full automation. Haptic shared control reduced speed fluctuations, braking instances, and jam lifetime relative to manual driving, but remained less effective than full automation in dissolving phantom traffic jams. Its distinguishing result concerns failure handling: during simulated silent automation failures, five out of 24 participants experienced a collision in the full-automation condition, whereas collisions did not occur with haptic shared control, and the minimum gap after failure was significantly larger. In JAD terms, this places human–automation coordination within the implementation space, rather than outside it (Koerten et al., 2022).

6. Practice-oriented formulations, adjacent control ideas, and research outlook

A recent practice-oriented formulation recasts police-car swerving behavior as a deployable JAD strategy. The “Single-Vehicle Two-Detector Jam-Absorption Driving” (SVDD-JAD) problem assumes only one dedicated vehicle and two stationary roadside detectors, one upstream and one downstream. Within that constrained sensing architecture, the strategy identifies five parameters that significantly affect operation: JAD speed $v_{\text{ini}}$0, inflow traffic speed $v_{\text{ini}}$1, wave width $v_{\text{ini}}$2, wave speed $v_{\text{ini}}$3, and in-wave traffic speed $v_{\text{ini}}$4. The maneuver geometry leads to the closed-form JAD-distance expression

$v_{\text{ini}}$5

with duration

$v_{\text{ini}}$6

Sensitivity analysis ranks the parameters from most to least influential as JAD speed, wave width, in-wave speed, inflow speed, and wave speed (He, 10 Feb 2026).

The practical claim of this formulation is narrower than that of the abstract JAD literature and therefore especially informative. It is designed for isolated stop-and-go waves rather than stationary, periodic congestion at classical bottlenecks. In a SUMO-based simulation, the strategy successfully suppresses the propagation of an isolated stop-and-go wave without triggering a secondary wave, and the required control region is reported as 1 to 10 km over 1–3 minutes. The parameter-estimation procedures rely on only two stationary detectors, and the implementation vehicle may be a manually driven police car rather than a fleet of CAVs. This suggests that a substantial part of the deployment barrier is architectural rather than dynamical: once the problem is reformulated around sparse sensing and a single intervention vehicle, the remaining difficulty is reliable parameter estimation and operational timing (He, 10 Feb 2026).

The broader review literature identifies the main open fronts for JAD as secondary waves, generalizability, traffic-state estimation and prediction, robustness to randomness, validation scenarios, and field tests and practical deployment. It also states that no formal large-scale field test of JAD has yet been reported, even though real-world analogs such as police-car swerving have been observed. In comparison with VSL, JAD requires less roadside infrastructure and offers finer vehicle-level control resolution, but it is more sensitive to parameter error, behavioral variability, and execution quality (He et al., 15 Apr 2025).

Related research on jam avoidance helps situate JAD within a wider control-theoretic landscape. In linear optimal-velocity models, adding a speed-difference term in autonomous car-following substantially increases stability, and autonomous models including speed difference are reported to be sufficient to maximise the stability (Tordeux et al., 2016). In a single-lane cellular-automata setting, jams can be dissolved by using local information about downstream vehicle velocities to induce defensive driving, thereby reducing inflow into the jam; if more than 80% of vehicles adopt the rule, jams dissolve successfully in both the $v_{\text{ini}}$7-vehicle and distance-based implementations (Lee et al., 2011). A plausible implication is that JAD should be understood not as an isolated maneuver class, but as one specific realization of a broader principle: local regulation of inflow and perturbation growth to destabilize the jam while preserving upstream stability.

The current state of the field therefore supports a precise but qualified conclusion. JAD can remove a target jam with remarkably sparse actuation, sometimes with a single absorbing vehicle, and it can do so in ways that improve travel time, fuel consumption, and collision risk under the appropriate operating conditions. At the same time, JAD is not inherently secondary-wave-free, not intrinsically robust to parameter misestimation, and not yet validated through formal large-scale field experiments. Its mature formulation is thus less a single algorithm than a family of vehicle-based wave-suppression strategies organized around three recurring technical themes: admissible absorbing velocities, accurate state and wave prediction, and stabilization of the upstream response (Nishi, 2019, He et al., 15 Apr 2025).