Multi-Objective Sparse Sensing with Ergodic Optimization

Ananya Rao, Howie Choset

TL;DR

Researchers at Carnegie Mellon University have developed a new framework (MO-SS-E) that allows robots to balance multiple competing goals—like finding scientific targets while avoiding steep slopes—even when they have a strictly limited "budget" for taking sensor measurements. By jointly optimizing the robot's path and its sensing schedule, the system achieves better area coverage with significantly fewer data points than traditional methods.

What Problem Does This Paper Address?

In robotic exploration (like a Mars rover mission), robots often have multiple objectives: they need to find interesting minerals, stay in the sun for solar power, and avoid hazardous terrain. Each of these objectives might require a different sensor.

The challenge is two-fold:

  1. Multi-Objective Balancing: How do you plan a path that satisfies three different "priority maps" simultaneously?
  2. Sparse Sensing: In the real world, taking a measurement isn't "free." It costs power, time, or data bandwidth. Most path-planning algorithms assume sensors are always on, but a resource-constrained robot needs to decide exactly when and where to click the shutter or fire the laser.

Existing "ergodic search" methods excel at driving robots to spend time in high-information areas, but they typically don't account for a limited sensing budget or the need to switch between different types of sensors on the fly.

Methods

The authors introduce the Multi-Objective Sparse Sensing Ergodic (MO-SS-E) metric. This approach builds on "ergodic optimization," which mathematically ensures that the time a robot spends in a region is proportional to the amount of information expected there.

Key technical components include:

  • The MO-SS-E Metric: A new cost function that combines three things: (1) how well the sensing locations match individual objective maps, (2) a penalty for taking too many samples (using $L_1$ regularization to promote "sparsity"), and (3) how well the overall trajectory covers a combined "master map."
  • Decision Variables ($\lambda$): The algorithm introduces a variable that acts as an "on/off" switch for each sensor at every point in time.
  • Relaxation and Projection: Since deciding "to sense or not to sense" is a hard binary choice (making the math difficult), they solve it as a continuous problem (0.0 to 1.0) and then "project" the results back to binary decisions that fit the budget.
  • Weighting via TOPSIS: To combine multiple objective maps without a human having to manually tune weights, they use the TOPSIS method to pick a "Pareto-optimal" balance.

Datasets

The framework was evaluated on two types of data:

  1. Synthetic Data: Randomized maps featuring Gaussian "blobs" representing areas of high interest.
  2. Real-World Data: Derived from satellite and aerial data of Cuprite, Nevada. They generated three specific objective maps:
  3. Entropy Map: Areas of high scientific uncertainty (using ASTER/AVIRIS data).
  4. Shade Map: Areas with high sunlight for solar-powered rovers.
  5. Slope Map: A risk map where high slopes are avoided (derived from Digital Elevation Models).

Key Insights and Ideas

  • Joint Optimization is Key: Instead of planning a path first and then deciding where to sample, doing both at once leads to much better performance. The robot "knows" it has to take a measurement soon, so it steers toward a high-value spot specifically for that measurement.
  • Heterogeneity Support: The math works for teams of different robots. For example, if Robot A has a "blue" sensor and Robot B has a "green" sensor, the framework coordinates them so they don't duplicate effort and cover all objectives efficiently.
  • Robustness to Low Budgets: The MO-SS-E approach shines most when the sensing budget is very low (e.g., only allowed to take measurements 10% of the time). It manages to keep the "ergodic error" low even as the number of samples drops.

Key Results

The authors compared MO-SS-E against two baselines: Uniform Sampling (sensing at fixed intervals) and a Probabilistic Heuristic.

  • Superior Coverage: In almost all tests (Single Agent, Homogeneous Team, and Heterogeneous Team), MO-SS-E achieved lower ergodic metrics (meaning better coverage) than the baselines.
  • Efficiency: On real-world data, MO-SS-E consistently outperformed the probabilistic heuristic at the 10% and 25% sensing budget levels.
  • Multi-Agent Scaling: The system successfully coordinated teams of up to 10 agents, demonstrating that the "sparsity" constraint helps distribute sensing tasks across a fleet effectively.

Limitations and Future Work

  • One Sensor at a Time: The current model assumes a robot can only use one sensor at a moment. In reality, some sensors can run in parallel, which the authors plan to address in "sensor fusion" updates.
  • The "A Priori" Assumption: The method requires "objective maps" to be known in advance. In true exploration, these maps are often built on the fly.
  • Sensor Noise: The study assumes perfect measurements. Future iterations will need to account for the fact that real-world sensors are noisy and might require repeat measurements in high-value areas.

Summary updated

Abstract

We consider a search problem where a robot has one or more types of sensors, each suited to detecting different types of targets or target information. Often, information in the form of a distribution of possible target locations, or locations of interest, may be available to guide the search. When multiple types of information exist, then a distribution for each type of information must also exist, thereby making the search problem that uses these distributions to guide the search a multi-objective one. In this paper, we consider a multi-objective search problem when the cost to use a sensor is limited. To this end, we leverage the ergodic metric, which drives agents to spend time in regions proportional to the expected amount of information there. We define the multi-objective sparse sensing ergodic (MO-SS-E) metric in order to optimize when and where each sensor measurement should be taken while planning trajectories that balance the multiple objectives. We observe that our approach maintains coverage performance as the number of samples taken considerably degrades. Further empirical results on different multi-agent problem setups demonstrate the applicability of our approach for both homogeneous and heterogeneous multi-agent teams.
BibTeX
@inproceedings{rao2023multi,
  author = {Ananya Rao and Howie Choset},
  title = {{Multi-Objective Sparse Sensing with Ergodic Optimization}},
  booktitle = {IEEE/RSJ International Conference on Intelligent Robots and Systems},
  year = {2023},
  doi = {10.1109/iros55552.2023.10341535},
}

Loading related content...

Figure