Integrated Longitudinal and Lateral Control for Autonomous Driving

You are reading version 1 of this direction, frozen 2026-10-04. The latest version is version 2.

Version 1 of 2329 papers365 conceptsfrozen 2026-10-04

For decades, the study of autonomous vehicle behavior has been largely one-dimensional. Most classical models, such as the Intelligent Driver Model (IDM) or standard Adaptive Cruise Control (ACC), focus exclusively on longitudinal control: the art of maintaining a safe speed and distance from the car directly ahead. While this "car-following" paradigm works for simple highway cruising, it fails to capture the complexity of real-world driving, where vehicles must constantly negotiate lane changes, merges, and intersections.

A growing body of research argues that the next leap in autonomy requires breaking down the wall between longitudinal speed control and lateral steering control. By integrating these two dimensions, researchers are moving from simple car-following to sophisticated, multi-lane maneuvering.

The Decoupling Problem

In traditional autonomous architectures, lateral and longitudinal dynamics are often treated as independent modules. A longitudinal controller manages the throttle and brakes to track a target speed, while a lateral controller manages the steering wheel to stay centered in a lane.

However, as Xiangru Xu, Jessy W. Grizzle, and colleagues point out in Correctness Guarantees for the Composition of Lane Keeping and Adaptive Cruise Control, these systems are physically coupled. A sharp turn introduces tire drag that slows the vehicle down, and high speeds significantly change how a vehicle responds to steering inputs. When these systems don't "talk" to each other, the vehicle can become unstable during aggressive maneuvers.

Beyond physics, decoupling limits intelligence. Jie Cheng 0008 and Yingbing Chen observe in PLUTO that even state-of-the-art imitation learning planners often struggle with lateral maneuvers like swerving or lane changes, despite being proficient at following a straight line. This "longitudinal bias" prevents agents from navigating dense urban environments where non-linear movement is a necessity, not an option.

From Car-Following to Integrated Maneuvering

To bridge this gap, researchers are developing frameworks that consider both axes of motion simultaneously. This integration typically happens through three main mechanisms:

1. Hierarchical Orchestration Instead of a single "do-it-all" model, frameworks like EnsembleFollower, developed by Xu Han, Xianda Chen, and others, use Reinforcement Learning (RL) to orchestrate a committee of models. By intelligently switching between rule-based longitudinal logic and data-driven lateral planning, the system achieves smoother, more human-like transitions during complex tasks like lane changes EnsembleFollower.

2. Dimensional Reduction (Virtual Rotation) One elegant way to handle the complexity of 2D space is to mathematically project it back into 1D. Meng Wang and Siyuan Gong proposed a "virtual rotation" approach for highway merging. By treating a merging ramp as a "virtual" lane behind the main traffic, they simplify a complex 2D steering problem into a 1D car-following scenario that accounts for both longitudinal spacing and lateral alignment Connected and Automated Vehicle Distributed Control for On-ramp Merging Scenario.

3. Unified State Representations Modern deep learning models are increasingly using Bird’s-Eye-View (BEV) maps to provide a unified spatial context. Raphael Trumpp and Abhinav Valada demonstrate that using BEV representations allows an agent to learn integrated maneuvers—managing steering and acceleration simultaneously—far more efficiently than models that treat them as separate tasks Efficient Learning of Urban Driving Policies Using Bird's-Eye-View State Representations.

Why Integration Matters for Traffic

The benefits of integrated control extend beyond the individual vehicle to the entire traffic ecosystem. In mixed-autonomy scenarios (where AVs drive alongside humans), longitudinal-only control has a major weakness: human drivers can simply bypass a "slow" autonomous traffic-smoother by changing lanes.

Research on "choreographing" mixed traffic shows that when autonomous vehicles are given lateral control (the ability to change lanes), they can effectively manage traffic across all lanes simultaneously. This prevents humans from bypassing regulation and can reduce average waiting times at intersections by over 50% Learning to Change: Choreographing Mixed Traffic Through Lateral Control and Hierarchical Reinforcement Learning.

Similarly, in platooning—where groups of vehicles travel closely together—integrating lateral steering is essential for resilience. As Hong Wang and Ehsan Hashemi argue, a platoon that only manages longitudinal distance will dissolve or collide when faced with a sudden obstacle requiring a swerve. A "holistic" controller that manages both axes ensures the platoon stays together even during emergency maneuvers A Holistic Robust Motion Controller Framework for Autonomous Platooning.

The Path Forward

The transition from 1D car-following to 2D integrated control represents a shift toward more "embodied" AI in transportation. By acknowledging the physical and strategic interdependencies between speed and steering, researchers are creating systems that don't just follow the car in front, but actively navigate the complex, multi-lane world around them. As Xinchen Zhou and Shidian Ma suggest, this integration is the key to making autonomous vehicles feel natural and predictable to the humans they share the road with A New Single Point Preview-Based Human-Like Driver Model on Urban Curved Roads.

Papers in this version

Showing 60 of 329 papers.

History & citationVersion 2 · 2026-10-04

Current · Permanent link

  1. Version 2 (latest) · 2026-10-04 · 167 papers
  2. Version 1 · 2026-10-04 · 329 papers
Cite this direction

Latest version (a living page):

@misc{lacuna_longitudinal_only_control_b42743d3,
  title = {longitudinal-only control},
  author = {{Lacuna}},
  howpublished = {Lacuna research direction},
  year = {2026},
  url = {https://lacuna.tiptreesystems.com/direction/longitudinal-only-control/rdr_b42743d33b48407d8e2c01d4e4245ba5},
  note = {Latest version of a living research direction; cite a version for a fixed text. Accessed 2026-11-20},
}

This exact version (version 1, frozen text):

@misc{lacuna_integrated_longitudinal_and_lateral_cont_b42743d3_v1,
  title = {Integrated Longitudinal and Lateral Control for Autonomous Driving},
  author = {{Lacuna}},
  howpublished = {Lacuna research direction},
  year = {2026},
  url = {https://lacuna.tiptreesystems.com/direction/integrated-longitudinal-and-lateral-control-for-autonomous-driving/rdr_b42743d33b48407d8e2c01d4e4245ba5/v/1},
  note = {Version 1, frozen 2026-10-04. Accessed 2026-11-20},
}