Optimizing Robotic Arm Obstacle Avoidance via Improved Random Tree Star (RRT)* and Deep Reinforcement Learning Coordination

Tianxuan Fu, Xing TangSymmetry 2025 · DOI

TL;DR

This paper introduces a novel collaborative framework that fuses an improved RRT algorithm with deep reinforcement learning (DRL) to tackle the critical challenge of dynamic obstacle avoidance for robotic arms in Industry 5.0. By synergistically combining RRT's efficient path exploration with DRL's adaptive learning, the framework significantly boosts success rates, reduces planning time, and enhances robustness in complex, real-world manufacturing scenarios.

What Problem Does This Paper Address?

The increasing adoption of Industry 5.0, with its emphasis on human-robot collaboration in smart manufacturing and intelligent logistics, demands highly efficient and reliable obstacle avoidance for robotic arms. However, existing methods fall short: traditional path planning algorithms like RRT struggle with dynamic environments, exhibiting path oscillations and poor real-time performance. Deep Reinforcement Learning (DRL) offers adaptability but suffers from low sample efficiency and high computational costs, making it unsuitable for industrial demands. This paper addresses the gap by proposing a hybrid approach that leverages the strengths of both RRT* and DRL to overcome these limitations, enabling robotic arms to navigate complex, unpredictable environments safely and efficiently.

Methods

The core of this research is a collaborative optimization framework that integrates an improved Rapidly-exploring Random Tree Star (RRT*) with Deep Reinforcement Learning (DRL).

  • Improved RRT*: The RRT* algorithm is enhanced with an adaptive sampling strategy that dynamically adjusts its Gaussian distribution based on obstacle prediction. This strategy intelligently focuses sampling on low-risk areas, reducing invalid sampling by 62% and improving computational efficiency. It also incorporates a dynamic reconnection mechanism for local path adjustments and a spatiotemporal attention module for better obstacle motion prediction.
  • Deep Reinforcement Learning (DRL): DRL is employed for its ability to adapt to environmental changes. In this framework, DRL acts as a global optimizer, enhancing dynamic obstacle prediction and providing residual torque corrections to the RRT*-generated paths. The DRL agent is trained using a composite reward function that balances task success, safety, and energy efficiency.
  • Collaborative Fusion: The RRT and DRL components are fused dynamically. Initially, RRT dominates, providing a feasible path. As the system learns, the DRL component's influence increases, allowing for more adaptive and optimized control. This is managed by a dynamic scheduling weight that decays over time.

Datasets

The paper does not explicitly mention specific public datasets used for training or evaluation. Instead, it describes the establishment of an experimental platform built upon ROS and Gazebo. This platform is designed to replicate physical properties and sensor noise characteristics of mechanical tubes in dynamic environments, serving as a simulation environment for testing and validation. The paper also discusses experiments conducted on a physical test platform using a UR5e cooperative manipulator, Intel NUC, RealSense D455 depth camera, and OptiTrack motion capture system, simulating real-world industrial scenarios.

Key Insights and Ideas

  • Synergistic Power of Hybridization: The most significant insight is that combining the geometric search capabilities of RRT with the adaptive learning of DRL creates a powerful synergy that overcomes the individual limitations of each approach. RRT provides a robust initial path, while DRL refines it and predicts dynamic obstacles more effectively.
  • Adaptive Sampling is Crucial: The improved RRT* with its adaptive sampling strategy significantly reduces wasted computation by focusing on relevant areas, a key factor in achieving real-time performance.
  • Dynamic Reconnection for Robustness: The dynamic reconnection mechanism is vital for handling unexpected obstacle movements, allowing the system to recover from path failures and maintain continuous operation.
  • Multi-Objective Optimization Pays Off: The composite reward function for DRL, balancing success rate, safety, and energy, is critical for achieving practical industrial performance, demonstrating that optimizing for multiple criteria is more effective than focusing on a single one.
  • Industry 5.0 Alignment: The framework directly addresses Industry 5.0's goals of safe, flexible, and sustainable human-robot collaboration by improving safety, reducing energy consumption, and enabling efficient production.

Key Results

  • Task Success Rate: The proposed framework achieved a task success rate of 93.8%, a significant improvement of 21.5% over traditional RRT*.
  • Path Length: The average path length was reduced to 1.97 m.
  • Energy Consumption: System energy consumption was 12.6 kWh.
  • Extreme Scenarios: In multi-robot collaboration, a 94.7% success rate was achieved.
  • Industrial Applications:
    • Automobile manufacturing assembly cycle time reduced to 8.4 s per task.
  • Yield rate improved to 98.7%.
  • Energy consumption reduced by 34%.
  • Human intervention reduced by 85.6%.
  • Planning Time: The planning time was reduced to 28.3 ms.
  • Obstacle Density Impact: The framework showed strong performance even at high obstacle densities, maintaining a success rate of 77.1% at 2.0 obstacles/m³.
  • Load Robustness: The framework demonstrated robustness under varying payloads, maintaining a success rate of 72.4% with a 10 kg load.

Limitations and Future Work

  • Limitations:
    • While the framework shows significant improvements, performance can degrade in extreme dynamic scenarios with very high obstacle densities or speeds.
  • Sensor noise and mechanical errors in physical implementations can lead to slight deviations from simulation results.
  • The paper acknowledges that the performance of existing methods in unstructured environments is still limited by uncertainties in sensor noise and dynamic obstacle movements.
  • Future Work:
    • Further integration of interpretable artificial intelligence and ethical design for human-robot interaction.
  • Exploration of multi-modal perception fusion in unstructured environments.
  • Hardware heterogeneity collaborative optimization.
  • Investigating human-robot interaction safety enhancement mechanisms.
  • Combining quantum computing and digital twin technologies to address real-time planning bottlenecks in high-dimensional configuration spaces.

Summary updated

Abstract

Driven by Industry 5.0, efficient obstacle avoidance of robotic arms in dynamic environments is a key bottleneck for human–robot collaboration in smart manufacturing. Traditional path planning methods such as Rapidly-exploring Random Tree and artificial potential field work stably in static settings but exhibit flaws including path oscillation and poor real-time performance under dynamic obstacles. Deep reinforcement learning adapts to environmental changes but is limited by low sample efficiency and high computational costs, failing industrial demands. This study proposes a collaborative framework integrating improved Rapidly-exploring Random Tree Star and Deep reinforcement learning. It uses Rapidly-exploring Random Tree Star to guide Deep reinforcement learning’s strategy exploration, reducing invalid sampling by 62%, and leverages Deep reinforcement learning’s global optimization to enhance dynamic obstacle prediction. The framework achieves a task success rate of 93.8%, surpassing traditional Rapidly-exploring Random Tree Star by 21.5%, with an average path length of 1.97 m and system energy consumption of 12.6 kWh. Experiments demonstrate superior performance in extreme dynamic scenarios, including a 94.7% success rate in multi-robot collaboration. Industrial cases confirm improvements in automobile manufacturing assembly cycle time to 8.4 s per task, yield rate to 98.7%, and reductions in energy consumption by 34% and human intervention by 85.6%, providing a reliable dynamic obstacle avoidance solution for Industry 5.0 applications.
BibTeX
@inproceedings{fu2025optimizing,
  author = {Tianxuan Fu and Xing Tang},
  title = {{Optimizing Robotic Arm Obstacle Avoidance via Improved Random Tree Star (RRT)* and Deep Reinforcement Learning Coordination}},
  booktitle = {Symmetry},
  year = {2025},
  doi = {10.3390/sym17122112},
}
Figure