[Paper Review] Kilombo: a Kilobot simulator to enable effective research in swarm robotics

Fredrik Jansson, Matthew Hartley+10 more|arXiv (Cornell University)|Nov 13, 2015

Modular Robots and Swarm IntelligenceEngineering5 references18 citations

TL;DR

Kilombo is a C-based simulator that enables direct execution of Kilobot code on both physical robots and simulations, eliminating the need for code translation and accelerating development. It achieves 100x real-time speedup for 1000-robot simulations, enabling high-throughput pre-screening of swarm algorithms with strong behavioral fidelity to physical robots.

ABSTRACT

The Kilobot is a widely used platform for investigation of swarm robotics. Physical Kilobots are slow moving and require frequent recalibration and charging, which significantly slows down the development cycle. Simulators can speed up the process of testing, exploring and hypothesis generation, but usually require time consuming and error-prone translation of code between simulator and robot. Moreover, code of different nature often obfuscates direct comparison, as well as determination of the cause of deviation, between simulator and actual robot swarm behaviour. To tackle these issues we have developed a C-based simulator that allows those working with Kilobots to use the same programme code in both the simulator and the physical robots. Use of our simulator, coined Kilombo, significantly simplifies and speeds up development, given that a simulation of 1000 robots can be run at a speed 100 times faster than real time on a desktop computer, making high-throughput pre-screening possible of potential algorithms that could lead to desired emergent behaviour. We argue that this strategy, here specifically developed for Kilobots, is of general importance for effective robot swarm research. The source code is freely available under the MIT license.

Motivation & Objective

• Address the slow and error-prone development cycle in swarm robotics caused by physical robot limitations such as recalibration, charging, and slow execution.
• Overcome the gap between simulator and physical robot behavior by enabling identical code to run on both platforms, avoiding translation errors.
• Provide a high-speed, efficient simulation environment that supports rapid hypothesis testing and algorithm screening for emergent swarm behaviors.
• Facilitate scalable and reproducible research by aligning simulation dynamics closely with real-world physical robot dynamics.
• Establish a reusable framework that can be extended to other low-cost, simple-robot platforms with similar capabilities.

Proposed method

• Develop a C-based simulator that compiles and executes the same source code used on physical Kilobots, ensuring API and behavioral compatibility.
• Implement a simulation loop that processes robot actions sequentially, treating loop execution as instantaneous to maximize speed while maintaining fidelity.
• Model robot movement, communication, and sensing with simplified physics to reduce computational overhead while preserving key stochastic behaviors.
• Ensure that messaging frequency and execution timing are modeled to reflect real-world constraints, minimizing discrepancies between simulation and physical behavior.
• Optimize performance by avoiding heavyweight physics engines and scripting language overhead, enabling 100x real-time simulation speed on standard desktop hardware.
• Design the simulator to be extensible, allowing adaptation to other robots by modifying only the robot API and environment representation components.

Experimental results

Research questions

• RQ1 Can a simulator be designed to run the same C code used on physical Kilobots, eliminating the need for code translation and reducing development errors?
• RQ2 To what extent does a high-speed simulator with simplified physics preserve the emergent behaviors observed in physical Kilobot swarms?
• RQ3 How does the simulation speed of Kilombo compare to real-time execution, and can it enable high-throughput pre-screening of swarm algorithms?
• RQ4 What trade-offs exist between simulation accuracy and performance, and how can they be balanced to maintain fidelity while enabling fast experimentation?
• RQ5 Can the simulator approach be generalized to other low-cost, simple-robot platforms with similar sensing and actuation capabilities?

Key findings

• Kilombo enables direct execution of Kilobot C code on both simulator and physical robots, eliminating translation errors and streamlining development.
• Simulations of 1000 Kilobots run at 100 times real-time speed on a standard desktop computer, enabling high-throughput algorithm screening.
• The simulator achieves strong behavioral fidelity to physical Kilobots, with deviations primarily arising from movement pattern constraints rather than fundamental algorithmic differences.
• The use of a lightweight, C-based simulation model avoids the performance overhead of scripting languages and heavyweight physics engines, significantly improving speed.
• The simulator’s design allows for accurate modeling of key dynamics such as communication delays and stochastic movement, even with simplified physics.
• The simulator is freely available under the MIT license and has been successfully used in real research projects, including the FET FoCAS SwarmOrgan Project.