ctm-dqn/docs/rule_vsl_baselines.md

Independent Rule-Based VSL Baselines

The project includes three deterministic non-learning VSL baselines. They are kept separate on purpose, so each one tests one traffic-control idea instead of mixing several rules into a single hand-crafted policy.

1. occ_rule_vsl

Local occupancy/speed hysteresis baseline.

The rule observes each controlled segment independently. When occupancy rises or local speed drops, the displayed speed limit is reduced in stages. When occupancy falls below a lower release threshold and speed recovers, the segment returns to the highest legal action. The lower release threshold creates hysteresis and avoids frequent switching around one threshold.

Literature basis:

• Smulders, S. (1990). Control of freeway traffic flow by variable speed signs. Transportation Research Part B: Methodological, 24(2), 111-132. doi:10.1016/0191-2615(90)90023-R.

2. bottleneck_rule_vsl

Downstream bottleneck pre-control baseline.

The rule does not react only to the current segment. For each controlled segment, it looks downstream over a short corridor window. If the downstream window contains high occupancy or low speed, the upstream segment is capped before traffic reaches the forming bottleneck. This isolates the idea of coordinated upstream flow moderation.

Literature basis:

• Hegyi, A., De Schutter, B., and Hellendoorn, J. (2005). Model predictive control for optimal coordination of ramp metering and variable speed limits. Transportation Research Part C: Emerging Technologies, 13(3), 185-209. doi:10.1016/j.trc.2004.08.001.

3. harmonization_rule_vsl

Speed-harmonization baseline.

The rule compares the measured speed of a segment with its immediate downstream segment. If the downstream segment is much slower, the upstream speed limit is reduced to smooth the speed transition. This isolates the safety-oriented speed harmonization idea without adding occupancy or bottleneck logic.

Literature basis:

• Allaby, P., Hellinga, B., and Bullock, M. (2007). Variable speed limits: Safety and operational impacts of a candidate control strategy for freeway applications. IEEE Transactions on Intelligent Transportation Systems, 8(4), 671-680. doi:10.1109/TITS.2007.908562.

Shared Display Constraints

All three baselines share only engineering display constraints:

• speed limits are selected from the same discrete action set as learning models;
• temporal changes are capped to avoid abrupt one-step jumps;
• adjacent segment changes are capped to avoid unrealistic spatial discontinuity.

These constraints are not additional control objectives; they only keep displayed VSL signs physically plausible and comparable across baselines.

Run Commands

Run one rule baseline:

uv run python -m training.run_model --model occ_rule_vsl
uv run python -m training.run_model --model bottleneck_rule_vsl
uv run python -m training.run_model --model harmonization_rule_vsl

Run all three:

uv run run_all_training.py --models occ_rule_vsl bottleneck_rule_vsl harmonization_rule_vsl

The default multi-model run now includes all three independent rule baselines.