A Hybrid SLAM and Object Recognition System for Pepper Robot (1903.00675v2)

Abstract: Humanoid robots are playing increasingly important roles in real-life tasks especially when it comes to indoor applications. Providing robust solutions for the tasks such as indoor environment mapping, self-localisation and object recognition are essential to make the robots to be more autonomous, hence, more human-like. The well-known Aldebaran service robot Pepper is a suitable candidate for achieving these goals. In this paper, a hybrid system combining Simultaneous Localisation and Mapping (SLAM) algorithm with object recognition is developed and tested with Pepper robot in real-world conditions for the first time. The ORB SLAM 2 algorithm was taken as a seminal work in our research. Then, an object recognition technique based on Scale-Invariant Feature Transform (SIFT) and Random Sample Consensus (RANSAC) was combined with SLAM to recognise and localise objects in the mapped indoor environment. The results of our experiments showed the system's applicability for the Pepper robot in real-world scenarios. Moreover, we made our source code available for the community at https://github.com/PaolaArdon/Salt-Pepper.

• The paper pioneers the integration of ORB-SLAM2 with a SIFT-RANSAC object recognition framework on Pepper, enabling robust simultaneous localization and mapping.
• The paper demonstrates a robust object recognition methodology that accurately identifies and maps objects even with limited sensor capabilities and challenging conditions.
• The paper validates the system in real-world experiments and provides open-source code to advance research in autonomous humanoid robotics.

A Hybrid SLAM and Object Recognition System for Pepper Robot

The paper "A Hybrid SLAM and Object Recognition System for Pepper Robot" presents a significant advancement in the field of robotic autonomy by integrating SLAM (Simultaneous Localization and Mapping) with object recognition capabilities in the Pepper humanoid robot. This research aims to enhance the operational scope and functionality of humanoid robots, particularly in indoor environments, by enabling them to perform autonomous navigation and object recognition tasks.

At the core of the paper is the implementation of a hybrid system that combines the ORB-SLAM2 algorithm with a robust object recognition framework utilizing SIFT (Scale Invariant Feature Transform) and RANSAC (Random Sample Consensus). This hybrid approach was specifically tailored and tested on the Pepper robot, which is equipped with limited sensing hardware, including RGB and depth cameras with modest resolution and frame rates.

Key Contributions

1. Visual SLAM Implementation on Pepper: The paper pioneers the adoption of a visual SLAM algorithm on the Pepper robot, extending its navigation capabilities beyond small and confined environments. This implementation allows the robot to autonomously map and localize within larger and more complex indoor spaces.
2. Robust Object Recognition Framework: The developed object recognition system leverages SIFT features combined with kd-tree and RANSAC for efficient and accurate object identification. This method provides resilience against object orientation and partial occlusions, offering consistent recognition performance even with a limited frame rate sensor.
3. Integration into a Unified Framework: The fusion of SLAM and object recognition creates a comprehensive system where detected objects are not only recognized but also spatially mapped in the robot’s generated environment model. This enables Pepper to maintain an updated map with annotated objects, facilitating better interaction with its surroundings.
4. Real-World Validation and Open Source Contribution: The paper validates the hybrid system in real-world conditions, showcasing successful mapping and object identification in dynamic environments. Additionally, the authors contribute to the robotics community by releasing their source code, which is accessible to enable further research and development in similar autonomous robotic applications.

Experimental Setup and Results

Experiments conducted highlight the robustness of the developed system in performing SLAM and object recognition concurrently. Key performance indicators include fast initialization and the ability to manage and update maps dynamically. The results demonstrated that object recognition achieved high accuracy and consistency, surpassing traditional methods such as Haar Cascades in terms of speed and reliability.

Furthermore, the enhancements made to the ORB-SLAM2 algorithm, including serialized map storage and fast vocabulary loading, significantly improved operational efficiency and the system's ability to resume previously interrupted mapping sessions seamlessly.

Implications and Future Directions

The integration of SLAM with object recognition in humanoid robots such as Pepper not only broadens their functional capabilities but also enhances their potential applications in home service robotics, healthcare, and elder care. The ability to autonomously navigate and interact with objects offers a foundation for tasks such as personalized assistance and environmental adaptation.

Looking forward, there are several avenues for further research and enhancement:

• Autonomous Path Planning: Implementation of path planning algorithms could enable Pepper to navigate autonomously towards specific objects or areas.
• Holistic Environment Interaction: Integration with manipulators for object retrieval and interaction could extend application scenarios.
• Improved Sensing and Computing Capabilities: Leveraging advancements in camera technology and onboard processing could refine map quality and system responsiveness.

In summary, this paper provides a well-defined methodology and experimental evidence for combining SLAM and object recognition within a humanoid robot framework, setting the stage for more advanced autonomous functionalities in service robots. The release of the implementation as open-source offers a valuable resource for continued innovation in the field.