Play Everywhere: A Temporal Logic based Game Environment Independent Approach for Playing Soccer with Robots

TL;DR

The paper tackles the challenge of making robot soccer robust to unstructured and changing environments by deriving rules and goals from environment semantics and executing them via a hierarchical soccer representation. It introduces a semantic map-based architecture that feeds a $PLTL_f$-based decision-making module into a $FOND$ planning framework, translating temporal goals into executable plans over PDDL domains. Three use-cases demonstrate progressively richer semantic inputs and temporal constraints, showing adaptive behavior from reaching a ball to safe, goal-oriented play on an SPL field. The approach offers a scalable, environment-aware alternative to fixed-rule or purely learning-based strategies, with potential applicability to RoboCup SPL and other dynamic tasks. Future work includes improving field element extraction through image segmentation to further enhance situational understanding.

Abstract

Robots playing soccer often rely on hard-coded behaviors that struggle to generalize when the game environment change. In this paper, we propose a temporal logic based approach that allows robots' behaviors and goals to adapt to the semantics of the environment. In particular, we present a hierarchical representation of soccer in which the robot selects the level of operation based on the perceived semantic characteristics of the environment, thus modifying dynamically the set of rules and goals to apply. The proposed approach enables the robot to operate in unstructured environments, just as it happens when humans go from soccer played on an official field to soccer played on a street. Three different use cases set in different scenarios are presented to demonstrate the effectiveness of the proposed approach.

Figures (5)

• Figure 1: Different scenarios where robots can play using the same architecture. The agent establishes the goal for the task from the semantics of the environment.
• Figure 2: Functional architecture. The sensory layer, here highlighted in green, is shown in Fig. \ref{['fig:sm_sensory_layer']}, while the conceptual layer, in red, is shown in Fig. \ref{['fig:sm_conceptual_layer']}.
• Figure 3: The detection modules in the sensory layer communicating with the conceptual layer, in red and shown in Fig. \ref{['fig:sm_conceptual_layer']}.
• Figure 4: Hierarchical architecture showing the relationships between the perceived objects and the semantic map in the conceptual layer.
• Figure 5: Policies generated from goals of increasing complexity. From left to right: $G_0 = O(isat\text{ } robot1\text{ } ballposition)$, $G_1 = G_0 \land O(goalscored)$, $G_2 = G_{SPL} = G_1 \land O(ballsafe\text{ } S\text{ } isat \text{ } robot1 \text{ } ballposition)$. Rectangular boxes contain fluents for the overlapped branch.