Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Learning Energy-Efficient Air--Ground Actuation for Hybrid Robots on Stair-Like Terrain

Jiaxing Li, Wen Tian, Xinhang Xu, Junbin Yuan, Sebastian Scherer, Muqing Cao

Abstract

Hybrid aerial--ground robots offer both traversability and endurance, but stair-like discontinuities create a trade-off: wheels alone often stall at edges, while flight is energy-hungry for small height gains. We propose an energy-aware reinforcement learning framework that trains a single continuous policy to coordinate propellers, wheels, and tilt servos without predefined aerial and ground modes. We train policies from proprioception and a local height scan in Isaac Lab with parallel environments, using hardware-calibrated thrust/power models so the reward penalizes true electrical energy. The learned policy discovers thrust-assisted driving that blends aerial thrust and ground traction. In simulation it achieves about 4 times lower energy than propeller-only control. We transfer the policy to a DoubleBee prototype on an 8cm gap-climbing task; it achieves 38% lower average power than a rule-based decoupled controller. These results show that efficient hybrid actuation can emerge from learning and deploy on hardware.

Learning Energy-Efficient Air--Ground Actuation for Hybrid Robots on Stair-Like Terrain

Abstract

Hybrid aerial--ground robots offer both traversability and endurance, but stair-like discontinuities create a trade-off: wheels alone often stall at edges, while flight is energy-hungry for small height gains. We propose an energy-aware reinforcement learning framework that trains a single continuous policy to coordinate propellers, wheels, and tilt servos without predefined aerial and ground modes. We train policies from proprioception and a local height scan in Isaac Lab with parallel environments, using hardware-calibrated thrust/power models so the reward penalizes true electrical energy. The learned policy discovers thrust-assisted driving that blends aerial thrust and ground traction. In simulation it achieves about 4 times lower energy than propeller-only control. We transfer the policy to a DoubleBee prototype on an 8cm gap-climbing task; it achieves 38% lower average power than a rule-based decoupled controller. These results show that efficient hybrid actuation can emerge from learning and deploy on hardware.

Paper Structure

This paper contains 34 sections, 29 equations, 6 figures, 1 table.

Table of Contents

  1. Introduction
  2. Related Work
  3. Hybrid aerial--ground robotic platforms
  4. Energy-aware planning and mode selection
  5. Learning for complex terrain traversal and multimodal locomotion
  6. System Overview
  7. Robot Dynamics with Hybrid Actuation
  8. Low-Level Control Interfaces and Policy Commands
  9. Decoupled Baseline Controller
  10. Simulation and Task Design
  11. Simulation Setup
  12. Task Description: Discontinuous Terrain Traversal
  13. Terrain Generation and Target Sampling
  14. Goal Direction Command
  15. Reinforcement Learning
  16. ...and 19 more sections

Figures (6)

  • Figure 1: Overview: we develop energy-aware reinforcement learning, trained in simulation and test in the real world.
  • Figure 2: Doublebee Actuators
  • Figure 3: Comparison between success rate and average consumption for hybrid mode, wheels only mode and propellers only mode.
  • Figure 4: Demonstration of emerged energy efficient behavior for traversing stair terrain, with the linear speed on robot's body frame's X/Z axis, pitch angle, thrust magnitude angle recorded.
  • Figure 5: Real-world RL policy execution on the stair-climbing task. Left to right: Phase I (approach), Phase II (climbing), Phase III (post-climb stabilization).
  • ...and 1 more figures