Abstract
In human-robot coexisting environment, both humans and robots are required to estimate each other's intention and act based on the estimation. In such an environment, it is not efficient to passively estimate the human intention. The objective of this study is to develop a robot that implicitly guides a human movement by its movement. We apply the social force model to prediction of human movements. Based on the prediction, the robot selects the appropriate motion to realize a desired movement of human.
Introduction
As mobile robots increasingly share spaces with humans, traditional collision avoidance strategies—which treat robots as passive observers—often prove inefficient. Such approaches frequently result in "deadlocks" or awkward maneuvers where both the human and robot stop or move in the same direction simultaneously. This paper proposes a proactive alternative: implicit guidance. Instead of merely reacting to human movement, the robot uses its own trajectory to "nudge" pedestrians toward a desired path, leveraging social norms and personal space as a steering mechanism.
Methodology
The Social Force Model (SFM)
The foundation of the proposed guidance system is the Social Force Model (SFM). This model treats human movement as a response to a set of virtual forces. A pedestrian $\alpha$ is influenced by an attractive force toward their goal ($F_\alpha^0$), repulsive forces from static obstacles ($F_{\alpha B}$), and repulsive forces from other agents, such as the robot $\beta$ ($wF_{\alpha\beta}$).
By calculating the resultant force $F_\alpha$, the robot can predict the human's future position ($r_\alpha$) over a short time interval $T$.
Action Quantization and Selection
To enable real-time decision-making, the robot's possible movements are discretized into a finite set of actions. The action space is divided into 8 directions (at $45^\circ$ intervals) and 4 speed levels: Stop, Low, Middle, and High. This results in 32 distinct motion patterns.
Optimization Loop
For each of the 32 possible actions, the robot simulates the human's response using the SFM. It then evaluates the predicted human position against a target "desired" position. The robot selects the specific action $RP(\theta, v)$ that minimizes the distance between the predicted and desired human locations, effectively "steering" the human through its own presence.
Key Insights
- Proactive Influence: The robot transitions from a passive obstacle to an active influencer, using its presence to manage traffic flow intuitively.
- Predictive Nudging: By simulating "what-if" scenarios across its action space, the robot can choose the movement that most effectively guides the human without explicit communication.
- Movement as Language: This approach relies on the "language of movement," which is often more natural in crowded environments than verbal or visual signals.
Conclusion and Future Work
This research establishes a mathematical framework for implicit human guidance. However, several challenges remain. The current model assumes a holonomic robot capable of instantaneous direction changes; future iterations must account for non-holonomic constraints common in car-like or differential-drive robots. Additionally, the model currently focuses on a single human-robot interaction; expanding this to complex social scenes with multiple pedestrians and attractive social forces is a key area for future development. Experimental validation in real-world environments is necessary to confirm the predictability of human responses to these social forces.
Summary updated