Reinforcement Learning for Reducing Congestion in Mixed Autonomy Highways
Bibliographic record
Abstract
Highway traffic dynamics are extremely complicated, due to the erratic behaviour of human drivers.Overreacting to avoid collisions, often by suddenly decelerating, causes waves of cars behind a driver to also slow down or stop, resulting in an emergent traffic jam.This results in many cars on the road travelling at a speed much lower than the theoretical maximum average speed of the road, and these delays create a significant cost to the economy.Inefficient driving behaviours have led to the increased interest in Autonomous Vehicles (AVs).There has also been a large amount of research on applying Machine Learning, and specifically Reinforcement Learning (RL), by using large amounts of data collected from the road, to ensure that individual AVs remain collision-free and more fuel efficient, and to maximize flow on roads that exclusively contain AVs.However, very little research has been done on mixed autonomy roads, especially in terms of improving traffic flow with AVs that are coordinated, but trained using their local observations of the road.This thesis focuses on assessing the impact of AVs on improving traffic flow in mixed autonomy collision-free highways.Roads were simulated using micro-traffic models, with adjustments designed to simulate a future world in which Automatic Braking Systems will be good enough to completely prevent rear-end and lane-change collisions on highways.We combined Deep Neural Networks with Reinforcement Learning (Deep Q Networks) to model the AVs, and this approach was tested on single and double lane highways.Among many other salient issues, our results show that replacing only 5% of human-drivers with AVs can, remarkably, improve average road speeds by up to 10% on single lane roads, up to 24% on double lane roads, and also significantly reduce the percentage of fully-stopped cars on average, for most situations.Furthermore, these trends, for the most part, continue to amplify as higher percentages of the human drivers are replaced with AVs.iii B.10 The dynamics-graph for the 3 lane (left) and 4 lane (right) proportional density change experiments. . . . . . . . . .
Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.
How this classification was reachedexpand
Full frame machine prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. The Gemma side is a direct model label for every work in the frame, read from the title-only record. The Codex side is a classifier learned from the 10,348 direct Codex labels and calibrated to design-weighted sample rates; fields without enough sample support carry no Codex call. Candidate is the union of the two sides; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels.
Distilled classifier scores by category (both heads)
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.001 | 0.002 |
| Meta-epidemiology (narrow) | 0.001 | 0.000 |
| Meta-epidemiology (broad) | 0.001 | 0.000 |
| Bibliometrics | 0.000 | 0.000 |
| Science and technology studies | 0.000 | 0.001 |
| Scholarly communication | 0.001 | 0.001 |
| Open science | 0.001 | 0.001 |
| Research integrity | 0.001 | 0.001 |
| Insufficient payload (model declined to judge) | 0.001 | 0.000 |
Machine scores (provisional)
The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.
Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.
score_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from itClassification
machine, unvalidatedMachine predicted; a candidate call from one source (direct Gemma or distilled Codex), not a consensus.
How this classification was reached, model by model and score by score, is at the end of the page under "How this classification was reached".