Reinforcement learning for end-to-end UAV slung-load navigation and obstacle avoidance.

Area of Science:

• Robotics
• Artificial Intelligence
• Control Systems

Background:

• Traditional control of Unmanned Aerial Vehicles (UAVs) with slung loads involves complex, separate systems for navigation, path planning, and obstacle avoidance.
• Existing Reinforcement Learning (RL) methods often require extensive pre-training and full-state observations, including potentially noisy load swing rates.

Purpose of the Study:

• To develop an integrated, end-to-end RL approach for UAV slung-load control that simplifies design and computation.
• To investigate a reduced observation space RL model (CompactRL-8) for improved efficiency and performance.
• To demonstrate the practical applicability and Sim2Real transfer capabilities of the proposed RL method.

Main Methods:

• An end-to-end RL framework was developed to unify navigation, path planning, and obstacle avoidance for UAVs with slung loads.
• A novel CompactRL-8 model was designed, utilizing only eight observations and excluding load swing rate measurements.
• The RL approach was validated using a detailed system model, achieving successful Sim2Real transfer without re-tuning.

Main Results:

• The CompactRL-8 model demonstrated a 58.79% increase in speed and a tenfold improvement in obstacle clearance compared to a full observation model.
• The proposed RL method outperformed state-of-the-art adaptive control techniques, showing an 8% enhancement in path efficiency and a fourfold increase in load swing stability.
• Successful Sim2Real transfer confirmed the robustness and practical applicability of the RL-based control strategy.

Conclusions:

• The unified RL approach offers a computationally efficient and effective solution for controlling UAVs with slung loads.
• The CompactRL-8 model provides superior performance and robustness, highlighting the benefits of a reduced observation space.
• This research paves the way for more reliable and efficient autonomous aerial systems for applications such as urban load transport.