Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Learning Hierarchical Control Systems for Autonomous Systems with Energy Constraints

Charlott Vallon, Mark Pustilnik, Alessandro Pinto, Francesco Borrelli

TL;DR

This work tackles the problem of operating autonomous systems under energy constraints by proposing a learning-enabled hierarchical control architecture that couples high-level energy-aware task planning with a low-level data-driven MPC. The method formalizes a two-tier SA/DM framework where upper-level estimates of route time and energy $(\hat{T}_{i,j}, \hat{E}_{i,j})$ are refined through learning, while the lower level converts plans into executable trajectories and adapts via iteration-based energy-model updates. Demonstrations on a two-vehicle electric delivery scenario show that iterative learning reduces conservatism and increases tasks completed within daily limits, though learning-rate choices critically affect stability and feasibility. Overall, the paper contributes a modular framework for energy-aware autonomous operation and highlights theoretical avenues for guaranteeing safety, feasibility, and real-time applicability of learning in hierarchical control.

Abstract

This paper focuses on the design of hierarchical control architectures for autonomous systems with energy constraints. We focus on systems where energy storage limitations and slow recharge rates drastically affect the way the autonomous systems are operated. Using examples from space robotics and public transportation, we motivate the need for formally designed learning hierarchical control systems. We propose a learning control architecture which incorporates learning mechanisms at various levels of the control hierarchy to improve performance and resource utilization. The proposed hierarchical control scheme relies on high-level energy-aware task planning and assignment, complemented by a low-level predictive control mechanism responsible for the autonomous execution of tasks, including motion control and energy management. Simulation examples show the benefits and the limitations of the proposed architecture when learning is used to obtain a more energy-efficient task allocation.

Learning Hierarchical Control Systems for Autonomous Systems with Energy Constraints

TL;DR

This work tackles the problem of operating autonomous systems under energy constraints by proposing a learning-enabled hierarchical control architecture that couples high-level energy-aware task planning with a low-level data-driven MPC. The method formalizes a two-tier SA/DM framework where upper-level estimates of route time and energy are refined through learning, while the lower level converts plans into executable trajectories and adapts via iteration-based energy-model updates. Demonstrations on a two-vehicle electric delivery scenario show that iterative learning reduces conservatism and increases tasks completed within daily limits, though learning-rate choices critically affect stability and feasibility. Overall, the paper contributes a modular framework for energy-aware autonomous operation and highlights theoretical avenues for guaranteeing safety, feasibility, and real-time applicability of learning in hierarchical control.

Abstract

This paper focuses on the design of hierarchical control architectures for autonomous systems with energy constraints. We focus on systems where energy storage limitations and slow recharge rates drastically affect the way the autonomous systems are operated. Using examples from space robotics and public transportation, we motivate the need for formally designed learning hierarchical control systems. We propose a learning control architecture which incorporates learning mechanisms at various levels of the control hierarchy to improve performance and resource utilization. The proposed hierarchical control scheme relies on high-level energy-aware task planning and assignment, complemented by a low-level predictive control mechanism responsible for the autonomous execution of tasks, including motion control and energy management. Simulation examples show the benefits and the limitations of the proposed architecture when learning is used to obtain a more energy-efficient task allocation.
Paper Structure (12 sections, 1 theorem, 28 equations, 5 figures)

This paper contains 12 sections, 1 theorem, 28 equations, 5 figures.

Table of Contents

  1. Introduction
  2. Literature Review
  3. Hierarchical planning systems
  4. Safe Reinforcement Learning and Iterative Learning Control
  5. Hierarchical Control Design
  6. Future Research
  7. Example
  8. Upper Level: Task Assignment
  9. Lower Level: Path Planning and Following
  10. Simulation Results
  11. Conclusions
  12. Appendix

Key Result

Proposition 1

The lower and upper bounds of a uniformly distributed variable $X \sim U(a,b)$ can be estimate with confidence $1-\alpha$ from $n$ independent identically distributed (i.i.d) samples by: where $M$ and $m$ are the maximal and minimal valued samples out of the $n$ independent samples.

Figures (5)

  • Figure 1: Real-world scenarios for Transit and Space Cyber-Physical Systems.
  • Figure 2: Framework with two levels
  • Figure 3: Planned tour on day one vs. day 20. The red node is the depot, blue nodes are customers, and green nodes are charging stations.
  • Figure 4: Planned vs. actual energy trajectory along tours. Tours planned with $P_E = 0.5$ initially quickly increase the number of visited customers, but become infeasible on day seven.
  • Figure 5: The planned (solid lines) and executed (dashed lines) time vs. state of charge trajectories along the edge (0, 20) over nine days.

Theorems & Definitions (2)

  • Proposition 1
  • proof