Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

A Rule-Based Behaviour Planner for Autonomous Driving

Bouchard Frederic, Sedwards Sean, Czarnecki Krzysztof

TL;DR

This paper tackles the need for transparent, reliable autonomous driving decision-making by proposing a two-layer rule-based behaviour planner that learns from expert decisions. The maneuver layer generates a set of candidate high-level behaviours from the perceived environment, which is then filtered by a conservativeness-based resolution to a single maneuver for the parameter layer, finalizing with a parameter-resolved driving policy: $RE(e) = \lambda_{par}(F_{R_{par}}(T_{par}(e)))$. Empirical results include a practical prototype deployed in a level-3 autonomous vehicle with a 110 km urban field test, achieving up to 300 Hz decision rates and about 98% autonomy comparable to a deep-learning baseline, while preserving explainability and debuggability. The authors provide a detailed learning/update cycle, including a backward-chaining coverage function and a knowledge-engineering workflow (discrepancy identification, misbehaviour diagnosis, knowledge extraction, and rule engineering) to incrementally grow and refine the rule base. The work demonstrates that rule-based approaches can achieve competitive autonomy in real-world urban ODDs and offers a tractable, auditable alternative to opaque end-to-end learning, with clear paths for extending the framework and integrating statistical robustness.

Abstract

Autonomous vehicles require highly sophisticated decision-making to determine their motion. This paper describes how such functionality can be achieved with a practical rule engine learned from expert driving decisions. We propose an algorithm to create and maintain a rule-based behaviour planner, using a two-layer rule-based theory. The first layer determines a set of feasible parametrized behaviours, given the perceived state of the environment. From these, a resolution function chooses the most conservative high-level maneuver. The second layer then reconciles the parameters into a single behaviour. To demonstrate the practicality of our approach, we report results of its implementation in a level-3 autonomous vehicle and its field test in an urban environment.

A Rule-Based Behaviour Planner for Autonomous Driving

TL;DR

This paper tackles the need for transparent, reliable autonomous driving decision-making by proposing a two-layer rule-based behaviour planner that learns from expert decisions. The maneuver layer generates a set of candidate high-level behaviours from the perceived environment, which is then filtered by a conservativeness-based resolution to a single maneuver for the parameter layer, finalizing with a parameter-resolved driving policy: . Empirical results include a practical prototype deployed in a level-3 autonomous vehicle with a 110 km urban field test, achieving up to 300 Hz decision rates and about 98% autonomy comparable to a deep-learning baseline, while preserving explainability and debuggability. The authors provide a detailed learning/update cycle, including a backward-chaining coverage function and a knowledge-engineering workflow (discrepancy identification, misbehaviour diagnosis, knowledge extraction, and rule engineering) to incrementally grow and refine the rule base. The work demonstrates that rule-based approaches can achieve competitive autonomy in real-world urban ODDs and offers a tractable, auditable alternative to opaque end-to-end learning, with clear paths for extending the framework and integrating statistical robustness.

Abstract

Autonomous vehicles require highly sophisticated decision-making to determine their motion. This paper describes how such functionality can be achieved with a practical rule engine learned from expert driving decisions. We propose an algorithm to create and maintain a rule-based behaviour planner, using a two-layer rule-based theory. The first layer determines a set of feasible parametrized behaviours, given the perceived state of the environment. From these, a resolution function chooses the most conservative high-level maneuver. The second layer then reconciles the parameters into a single behaviour. To demonstrate the practicality of our approach, we report results of its implementation in a level-3 autonomous vehicle and its field test in an urban environment.
Paper Structure (16 sections, 16 equations, 6 figures, 2 algorithms)

This paper contains 16 sections, 16 equations, 6 figures, 2 algorithms.

Table of Contents

  1. Introduction
  2. Rule Engine
  3. Layers and Rules
  4. Resolving a Single Behaviour
  5. Inference Example
  6. Learning and Maintaining the Theory
  7. Rule Engine Update Algorithm
  8. Rule and Training Set Development
  9. Discrepancy Identification
  10. Misbehaviour Diagnosis
  11. Knowledge Extraction
  12. Rule Engineering
  13. Experimental Results
  14. Driving Policy
  15. Field Test
  16. ...and 1 more sections

Figures (6)

  • Figure 1: Motion planning architecture of our autonomous vehicle
  • Figure 2: Diagrammatic representation of two-layer rule engine
  • Figure 3: Example scene: autonomous vehicle approaches intersection with crosswalk
  • Figure 4: Knowledge engineering cycle
  • Figure 5: Maneuver rule distribution
  • ...and 1 more figures