Source-linked AI summary
Tuning ROS 2 for Energy-Efficient Navigation: Empirical Insights from Costmap 2D Configurations
Michel Albonico, Andreas Wortmann, Ivano Malavolta
TL;DR
ROS 2’s configurable packages can yield suboptimal settings, affecting navigation performance and energy use. The paper tests Costmap 2D configurations in controlled warehouse experiments and finds that tuning matters across layouts: some settings reduce energy use, while others impair navigation and increase consumption.
Problem
ROS 2’s broad configuration space can produce suboptimal settings that affect performance and energy efficiency, making workload-specific configuration an important question for mobile-robot navigation.
Source: §I · p. 1Method
The study evaluates Costmap 2D configurations through controlled navigation experiments, measuring energy, resource use, and performance across workload conditions.
Source: §I · p. 1, §V · p. 2, §VI · p. 3Results
Across small and large warehouse layouts, moderate resolution, low update frequencies, and reduced reliance on inflated or unknown-space layers were associated with lower energy use; rough resolutions and excessive inflation led to navigation problems and higher consumption.
Source: §I · p. 1Takeaways & Limitations
Costmap 2D parameter tuning, particularly resolution, update frequency, and inflation settings, had a stronger impact on energy consumption than plugin combinations in the tested ROS 2 navigation experiments.
Source: §IX · p. 8
Abstract
from arXiv · showhide
Robots are increasingly used in diverse application areas, where autonomous navigation plays a central role. As these systems become more widespread, improving their energy efficiency is critical to extending operational time and reducing environmental impact. The Robot Operating System (ROS) is a widely adopted middleware for robotics, offering a rich set of configurable packages. However, this flexibility can result in suboptimal software configurations in dynamic environments, negatively affecting both performance and energy consumption. This paper investigates the impact of ROS 2 package reconfigurations on the energy efficiency of mobile robot navigation. We conduct a controlled experiment in two warehouse-like scenarios (small and large) with varying obstacle layouts and Costmap 2D configurations (essential to the Nav2 stack). Through repeated trials, we measure energy usage, power profile, CPU load, memory consumption, and navigation performance. Results show that configurations must be carefully chosen for the specific robotic environment, and we were able to identify critical settings that lead to good and poor performance and energy consumption.
I. INTRODUCTION
The paper examines how ROS 2 Costmap 2D reconfiguration affects navigation energy efficiency, motivated by the difficulty of choosing effective settings from a large configuration space. Its experiments identify settings associated with better or worse energy and navigation performance.
I. INTRODUCTION
- Energy efficiency matters for extending robot operating time and minimizing environmental impact, while software components can themselves be a major source of energy consumption.ROS configuration flexibility in dynamic environments can lead to suboptimal settings that affect performance and energy efficiency.Source: §I · p. 1
- The study empirically assesses ROS package reconfiguration for mobile-robot navigation, proposing alternative Costmap 2D configurations and testing them in controlled warehouse experiments with repeated runs.The experiments vary navigation goal, Costmap 2D configuration, obstacle arrangement, and warehouse layout while keeping the robot model and other ROS package settings constant.Source: §I · p. 1
- Moderate resolution, low update frequencies, and reduced reliance on inflated or unknown-space layers were associated with lower energy usage, while rough resolutions and excessive inflation led to failures, longer paths, or corrective behaviors.The reported patterns held across small and large warehouse layouts.Source: §I · p. 1
II. ROS CONFIGURATION
- ROS packages expose configurable parameters, but the resulting configuration space can become large and produce suboptimal settings that compromise performance and resource use.The paper defines the space as the Cartesian product of admissible parameter values across the system’s packages.Source: §II · pp. 1–2
III. RELATED WORK
Prior work studies ROS energy use and software reconfiguration, but the paper distinguishes its focus on measuring Nav2 configuration impacts. Its warehouse mission models travel from a charging station to a fixed goal across two layouts.
III. RELATED WORK
- Prior ROS energy studies examine architectural tactics, programming languages, edge offloading, and SLAM packages, but not energy-oriented reconfiguration of Nav2 packages.Source: §III · p. 2
- Related reconfiguration work covers ROS architecture, customization, and misconfigurations, but does not measure the energy or performance impact of robotic software configurations.Source: §III · p. 2
IV. ROBOTIC MISSION
- The mission models a robot leaving a hypothetical charging station for a fixed goal across small and large warehouse maps, with dynamically added obstacles that do not block the route.The small map measures 3 × 4 m; the large map is approximately 10 times larger, and the planned paths are approximately 6 m and 18 m, respectively.Source: §IV · p. 2
V. STUDY DESIGN
The study uses a Goal Question Metric framework to examine Costmap 2D’s relationship to energy efficiency in ROS 2 mobile robots. Its design separates controlled conditions, configuration and workload factors, and navigation, resource, and energy outcomes.
V. STUDY DESIGN
- Following the Goal Question Metric framework, the study asks how Costmap 2D configurations affect energy consumption, which parameters matter, how workload interacts with configuration, and how findings can inform practice.Source: §V · p. 2
- The design holds the robot model, Nav2 installation, hardware and virtual machines, initial position, and navigation goal constant across runs.Table I summarizes the study metrics.Source: §V · p. 2, Table I · p. 2
- The independent variables are Costmap 2D configurations, obstacle density, and warehouse layout, while dependent measures cover navigation performance, resource usage, and energy efficiency.The configurations were validated by two experts.Source: §V · p. 3
VI. EXPERIMENT EXECUTION
The experiment narrows a configuration space too large for exhaustive testing, then runs representative Costmap 2D settings under controlled simulation and profiles energy, resources, and navigation. Execution took about 33 hours for the small map and 3.14 days for the large map.
VI. EXPERIMENT EXECUTION
- The experiment considered 18 Costmap 2D parameters, with continuous parameters restricted to representative finite values validated by two experts.The parameter values used in the experiment are summarized in Table II.Source: §VI · p. 3, Table II · p. 4
- Pairwise sampling reduced a space of more than 30 million parameter combinations to 20 representative configurations, using expert-informed constraints.The approach targets pairwise parameter interactions to make experimentation feasible.Source: §VI · p. 3
- Robot Runner orchestrated simulation, Nav2 execution, profiling, and cleanup, while profilers measured machine and controller/planner power, CPU, memory, and navigation performance.Runs used simulator time and a 60-second cooldown between trials.Source: §VI · p. 3
VII. RESULTS
- The small-map experiment lasted approximately 33 hours, compared with approximately 3.14 days for the large-map experiment.The runtime calculation accounts for run duration, cooldown, configuration count, obstacle arrangements, and repetitions.Source: §VII · p. 4
A. Power and Energy Usage
Costmap configurations affected both node-level and whole-machine power and energy. The configurations with lower node-level energy did not all show the same advantage in whole-machine energy.
A. Power and Energy Usage
- Only c2, c4, and c12 achieve lower overall energy for the controller and planner nodes.For c4, high planner power is offset by short navigation time.Source: A. Power and Energy Usage · p. 4
- For the whole machine, most configurations use less power than default, but only c3 has higher energy efficiency; c4 and c12 are near default.c9 and c13 are exceptions to the lower-power pattern.Source: A. Power and Energy Usage · p. 4
B. Resource Usage
Costmap configurations vary in resource use and navigation outcomes. Shorter routes do not always mean shorter navigation time, while some configurations combine reliable completion with substantial recovery behavior.
B. Resource Usage
- Memory use for c3, c4, and c12 is consistently lower than default, while CPU usage varies across configurations.Memory is not directly included in the energy measurements; CPU activity and power do not align consistently because processor mechanisms affect power use.Source: B. Resource Usage · p. 4
C. Navigation Performance
- Most configurations produce paths at least one meter shorter than default, but c9 more than doubles navigation time and total energy despite a route over one meter shorter.The reduced energy for c3, c4, and c12 is primarily explained by shorter navigation times.Source: C. Navigation Performance · pp. 4–5
- In the small map, c6, c8, c12, c14, c16, c17, and c18 succeed in all 20 repetitions, while c5, c10, c11, c13, and c19 stay below 80%.Other configurations exceed 80% success; their few failures are attributed to unexpected experiment-orchestration conditions, not configuration flaws.Source: C. Navigation Performance · p. 5
- c6 succeeds in every run despite many recoveries, whereas c11 succeeds in only 60% despite requiring none.The contrast suggests c6 relies on corrective actions, while c11's failures could not be mitigated by recovery strategies.Source: C. Navigation Performance · p. 5
D. Large Map Results
In the large map, resource use and navigation behavior differ from the small-map setting. Wider free space is associated with lower controller power, while excessive inflation can lead to long, oscillatory routes near obstacles.
D. Large Map Results
- Despite a map approximately 10× larger, mean CPU cycles and measured power are lower, driven mainly by controller power falling to roughly half.Wider corridors may reduce controller avoidance maneuvers and micro-corrections; planner power decreases only modestly.Source: D. Large Map Results · p. 5
- Memory use also decreases more modestly; high usage in c2, c17, c18, and c19 occurs with broader inflation and unknown-space tracking alongside the voxel plugin.In the small map, c2's higher cost_scaling_factor narrows the inflated area and reduces cells stored and updated.Source: D. Large Map Results · p. 5
- Navigation path length and time approximately triple in the large map, while excessive inflation can steer the robot toward obstacles instead of open space.In c8, the replanned route detours through narrow passages and becomes unnecessarily long and oscillatory.Source: D. Large Map Results · pp. 5–6
E. Statistical Analysis
The analysis tests configuration and obstacle-count effects on node power and overall energy, and compares measurements between map sizes. Because normality was often rejected, especially for small-map experiments, non-parametric tests were used.
E. Statistical Analysis
- For both maps, the analysis examined configuration and obstacle-count effects on controller and planner power and overall energy, and compared measurements between maps.Shapiro–Wilk tests often rejected normality, particularly in small-map experiments, so the study used non-parametric tests.Source: E. Statistical Analysis · p. 6
1) Effects of configurations:
Costmap 2D configuration significantly affected power and energy in both maps, while obstacle arrangements had no significant effects within the tested ranges. Map size also corresponded to marked differences in controller power and overall energy.
1) Effects of configurations:
- Configuration significantly affected controller power, planner power, and overall energy in both maps, with numerous significant pairwise differences among configurations.For the small map, overall energy was H=222.469, p=1.102×10−36; for the large map, overall energy was H=188.450, p=6.652×10−30.Source: 1) Effects of configurations · p. 6
2) Effects of Obstacles:
- Obstacle arrangements had no statistically significant effect on controller power, planner power, or overall energy in either map.Within the tested ranges, obstacle density had limited impact compared with configuration choices.Source: 2) Effects of Obstacles · p. 6
3) Cross-Run Comparisons (Small vs. Large Map):
- Controller power and overall energy differed strongly between maps, while planner power also differed but to a lesser extent.A two-way ANOVA found significant map effects for controller, planner, and overall energy, but no significant obstacle main effect or map-by-obstacle interaction.Source: 3) Cross-Run Comparisons (Small vs.… · p. 6
VIII. DISCUSSION
The discussion identifies configuration tuning as a stronger influence on navigation energy than plugin combinations, while emphasizing parameter interactions and environmental conditions. It recommends validating settings under realistic workloads and notes limits in the tested stack and robot setup.
A. Answering the Research Questions
- Configurations c3, c4, and c12 used less energy than the default, with shorter navigation times, lower node power, and reduced memory usage.The paper notes that memory was not directly captured in its power and energy measurements.Source: A. Answering the Research Questions · p. 6
- Parameter tuning had greater influence than plugin combinations; 0.1 m resolution, update frequency, and inflation and unknownspace settings were key factors.A 0.2 m resolution repeatedly caused unsuccessful navigation, while 1 Hz update frequency increased planner power through more intensive replanning.Source: A. Answering the Research Questions · p. 7
- Map size and layout shaped energy and performance: wider corridors reduced controller micro-corrections, whereas narrow environments increased avoidance and replanning demands.Obstacle arrangements were not statistically significant in these experiments, but the discussion says denser, more dynamic obstacle setups warrant further study.Source: A. Answering the Research Questions · p. 7
B. Practical Observations
- The most energy-efficient configurations shared shorter navigation times, lower planner or controller power, and no recovery behaviors.Their common pattern included 0.1 resolution, disabled inflate_around_unknown, and update_frequency of at least 5 Hz; 0.2 resolution was associated with unsuccessful navigation.Source: B. Practical Observations · p. 7
- Navigation time was a major determinant of total energy, and inflation_radius and track_unknown_space could restrict motion and increase travel time.The discussion also attributes accumulated pauses in c9 relative to c3 to transform_tolerance=0.5 versus 0.1.Source: B. Practical Observations · p. 7
- The evaluation covers Nav2 Costmap 2D on one robot model with fixed sensors and hardware, limiting evidence beyond that setup.The discussion calls for replication across robot platforms and additional ROS 2 packages; it also recommends runtime reconfiguration for real deployments.Source: B. Practical Observations · p. 8
IX. CONCLUSION
The conclusion reports that Costmap 2D configuration significantly influences energy efficiency in mobile robot navigation. It highlights parameter tuning, particularly resolution, update frequency, and inflation settings, and reports that c3, c4, and c12 outperformed the default setup.
IX. CONCLUSION
- ROS 2 Costmap 2D configurations significantly influence the energy efficiency of mobile robot navigation.The conclusion bases this finding on controlled experiments across different maps and obstacle layouts.Source: §IX · p. 8
- Resolution, update frequency, and inflation settings had stronger impact on energy consumption than plugin combinations.Source: §IX · p. 8
- c3, c4, and c12 outperformed the default setup by reducing navigation time, controller and planner power demands, and memory usage.Source: §IX · p. 8
- The conclusion identifies deliberate configuration tuning as essential to efficient navigation in ROS 2 systems.Source: §IX · p. 8