Reasoning for Moving Blocks Problem: Formal Representation and Implementation

Przemyslaw Andrzej WalegaarXiv 2013 · arXiv

Abstract

This paper presents a hybrid reasoning framework designed to address uncertainty in robotic planning tasks. By integrating Qualitative Reasoning with Probabilistic Functions within the Situation Calculus framework, the proposed system enables robots to perform complex block-moving tasks using symbolic labels rather than precise numerical coordinates. The implementation in Prolog demonstrates the effectiveness of this approach in simulated environments, highlighting its scalability and robustness in handling varying degrees of belief.

Introduction

In classical robotics, planning typically relies on precise numerical data. However, real-world applications are often hindered by noisy sensors and incomplete information. While Probabilistic Reasoning is effective for managing uncertainty, it is computationally intensive. Conversely, Qualitative Reasoning provides human-like "common sense" but often lacks the precision required for physical execution. This research bridges this gap by introducing a formal representation for the "Moving Blocks Problem," where a robot must transition from an initial shuffled state to a goal state without exact block counts or real-time feedback.

The simulated "Moving Blocks" environment featuring a robot and five columns of blocks.

Formal Framework: Situation Calculus

The system is built upon Situation Calculus, a logic-based language for reasoning about actions and change. The world is represented through:

  • Actions: Discrete operations such as move(n, n').
  • Situations: Sequences of actions that represent the history of the world state.
  • Fluents: Functions describing the state of the world, such as column height.

Hybrid Reasoning and Implementation

The core innovation lies in the use of functional fluents that return a "degree of belief" (ranging from 0 to 1) instead of simple integers. This approach utilizes triangular membership functions inspired by Fuzzy Logic to map numerical values to qualitative labels: Zero, Small, Medium, and Large.

Triangular membership functions mapping numerical block counts to qualitative labels (Zero, Small, Medium, Large).

As blocks are moved, these belief values shift numerically, while the high-level planning remains qualitative. The system is implemented in Prolog using the IndiGolog agent architecture, allowing for the formalization of "Causal Laws" that define how actions affect the robot's internal beliefs.

Evaluation and Results

The framework was evaluated in a simulated environment with five columns of cubic blocks. Testing focused on three scenarios: well-established initial states, borderline cases, and failure cases.

  • Success in Clear Cases: The robot achieved a 100% success rate when block counts were centered within a qualitative range (e.g., 7 blocks for 'Medium').
  • Borderline Performance: The system remained effective even at the edges of ranges (e.g., 8 blocks), successfully transitioning belief degrees.
  • Qualitative Failure: Failures occurred when initial qualitative assignments were ambiguous relative to the goal, leading the robot to stop prematurely.

Conclusion and Future Work

The hybrid approach offers significant scalability; increasing reasoning precision only requires adding more qualitative labels rather than restructuring the underlying logic. However, the "Borderline Problem" remains a challenge where persistent incorrect beliefs can prevent goal completion. Future work will focus on extending this logic to moving objects, physical robot implementation, and integrating the ExpCog framework for improved sensor noise filtering.

Summary updated

BibTeX
@misc{walega2013reasoning,
  author = {Przemyslaw Andrzej Walega},
  title = {{Reasoning for Moving Blocks Problem: Formal Representation and Implementation}},
  year = {2013},
  eprint = {1307.7405},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/1307.7405},
}
Figure