J. Taery Kim,Wenhao Yu,Yash Kothari,Jie Tan,Greg Turk,Sehoon Ha
Abstract:This paper explores the principles for transforming a quadrupedal robot into a guide robot for individuals with visual impairments. A guide robot has great potential to resolve the limited availability of guide animals that are accessible to only two to three percent of the potential blind or visually impaired (BVI) users. To build a successful guide robot, our paper explores three key topics: (1) formalizing the navigation mechanism of a guide dog and a human, (2) developing a data-driven model of their interaction, and (3) improving user safety. First, we formalize the wayfinding task of the human-guide robot team using Markov Decision Processes based on the literature and interviews. Then we collect real human-robot interaction data from three visually impaired and six sighted people and develop an interaction model called the ``Delayed Harness'' to effectively simulate the navigation behaviors of the team. Additionally, we introduce an action shielding mechanism to enhance user safety by predicting and filtering out dangerous actions. We evaluate the developed interaction model and the safety mechanism in simulation, which greatly reduce the prediction errors and the number of collisions, respectively. We also demonstrate the integrated system on a quadrupedal robot with a rigid harness, by guiding users over $100+$~m trajectories.
What problem does this paper attempt to address?
The main problem that this paper attempts to solve is to transform quadruped robots into guide robots for the visually impaired. Specifically, the paper explores three key topics:
1. **Formalization of the Navigation Mechanism**: The paper studies the navigation patterns between humans and guide dogs and uses Markov Decision Processes (MDPs) to establish a formal navigation task. This helps to develop a more natural and comfortable user interface to meet the needs of visually impaired users.
2. **Development of the Human - Robot Interaction Model**: The paper collects real - human - robot interaction data and develops a concise interaction model called "Delayed Harness". This model can more accurately predict the navigation behavior of the team, thereby improving navigation performance.
3. **Enhancement of User Safety**: The paper introduces an action shielding mechanism. By predicting and filtering dangerous actions, it improves the safety of visually impaired users. This mechanism has been tested in both simulated and real - world environments and has significantly reduced the number of collisions.
### Paper Background
Currently, guide dogs play an important role in the lives of the visually impaired, but their supply is limited, and only about 2 - 3% of potential users can benefit. Therefore, researchers have explored various assistive technologies, such as smart canes, belts, glasses, audio / tactile systems, and mobile robots. In recent years, with the development of quadruped robots, these robots have become a new research hotspot because of their autonomy and mobility.
### Main Contributions
1. **Formalization of the Navigation Task**: The paper proposes a formal definition of the navigation task for human - guide - dog teams.
2. **Development of the Interaction Model**: It collects human - robot interaction data, develops a concise interaction model, and open - sources these data to promote research in the robotics community.
3. **Enhancement of User Safety**: By introducing the action shielding mechanism, it improves the safety of visually impaired users.
4. **Hardware Demonstration**: A complete system is demonstrated on a quadruped robot, which can effectively navigate on a trajectory of more than 100 meters.
### Experimental Results
- **Accuracy of the Interaction Model**: By comparing with the fixed harness and rotating rod models, the higher accuracy of the "Delayed Harness" model is proven.
- **Effectiveness of Action Shielding**: In both ideal and noisy environments, the action shielding mechanism significantly reduces the number of collisions.
- **Practical Application**: In a real - world indoor environment, the robot successfully guides users to complete multiple long - distance tasks, demonstrating the robustness and practicality of the system.
### Limitations
Although the methods proposed in the paper have made significant progress, there is still room for improvement:
- Basic verbal instructions need to be extended to cover more complex commands and non - verbal signals.
- The current "Delayed Harness" model performs poorly in handling some special scenarios (such as when the user pauses chatting), and the data set needs to be expanded in the future.
- Although the action shielding mechanism is effective, it may still fail in some cases, and sensors and safety mechanisms need to be further optimized.
- Evaluation needs to be carried out in a larger group of visually impaired users to collect feedback and further improve the system.
In general, this paper provides an important theoretical basis and technical support for transforming quadruped robots into guide robots, laying the foundation for future related research.