Efficient and Trustworthy Social Navigation Via Explicit and Implicit Robot-Human Communication
Yuhang Che, Allison M. Okamura, Fellow, IEEE, and Dorsa Sadigh, Member, IEEE
Here, we are going to review the paper published by the above authors
This is part of my academic assignment in my graduate program. After reading the paper multiple times, I have come up with the following questions to answer related to this paper.
Paper Link: https://arxiv.org/abs/1810.11556
1. What is the problem being addressed?
In this paper, the authors discuss how usual robots don’t fully understand how humans interact socially. Generally, robots are trained to avoid obstacles and follow paths to achieve their desired target. However, this does not capture the nuanced interactions of humans, such as eye contact, body language, and gestures. Humans use these interactions to avoid bumping into each other. For robots to fit into human environments like offices and homes, they need to understand these interactions. In this paper, they addressed that avoiding collisions isn’t enough for a robot; it also needs to communicate how humans communicate, like signaling its next move, etc.
2. Why is that a problem worth solving?
The problem is worth solving because it directly improves the safety of robots sharing environments with humans. Humans, understanding these subtle interactions, mostly avoid bumping into each other. If robots are only programmed to avoid specific obstacles, this may lead to many collisions. For example, human movements can be unpredictable, and these movements cannot always be identified as obstacles by robots in certain cases. By addressing these problems, we can expand the applications for robots and enable them to behave more like humans.
3. How does the paper solve/address it?
The paper solves the problem of social interaction in robots by using implicit and explicit communications. Implicit includes eye contact, body language, etc., and explicit includes spoken language, audio signals, etc. The paper approaches the solution by:
- Developing a predictive model of human navigation behavior in response to both implicit and explicit robot communication.
- Creating an interactive planning algorithm based on the human model that enables a robot to clearly communicate its intentions.
- Implementing the proposed algorithm on a physical mobile robot platform. This study incorporates a wearable haptic device which can be used to convey the robot’s intentions.
This is validated in a controlled environment with humans and robots. They made both cross paths and observed that the planner can effectively convey the robot’s intentions.
4. What are the limitations of the presented ideas? Weak points? Untenable assumptions?
Limitations:
1) The entire study was conducted and observed in a controlled environment, where the authors assumed that humans would not interrupt the robots. However, implementing such methods in complex environments is very difficult. This may pose many challenges, as human behavior is unpredictable. Also, implementing such advanced systems may increase the resources and cost of the system, making it difficult to implement in real-world environments.
2) Sampling human behavior data is very difficult. This predictive model may not be effective enough to capture a wide variety of human behaviors.
5. What did you like about the paper?
1) I liked the way signs were incorporated into the robots, similar to how humans mostly communicate, whether through driving or walking. Humans understand the intentions of others by observing their approach. Implementing this innovative approach to social navigation is forward-thinking.
2) Although wearable haptic feedback is difficult to implement in the real world, using this device for explicit communication in a controlled environment is a creative solution. I liked how the author used it to convey explicit information.
6. Which safety issues related to AI systems are addressed in this paper? Please use the dimensions of AI safety presented in this Figure within the blue boxes (numbered 1–5) and justify your answer.
The paper addresses below two safety dimension:
Intent vs Specification: The paper deals this challenge by ensuring humans intent are aligns with robot’s behavior.For example, Imagine a user intends for a robot to follow user closely without colliding. If the robot is too cautious and keeps a distance , then the specification of following closely isn’t matched with user’s intent. The paper addresses this by modeling explicit and implicit communications, so that robot understands to stay close without the risk of collision, this aligns with user’s intent.
Computed Behavior vs Real Outcome: By the implementation of the computed algorithm on the physical mobile robot platform and testing it with human users in controlled environment , the paper evaluates how closely the computed behavior(Planned path & communications) matches the real outcome in terms of safety.
