Research map / Probabilistic and scientific modeling / Dynamical systems
Optimal Sensor Scheduling and Selection for Continuous-Discrete Kalman Filtering with Auxiliary Dynamics
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Derives a differentiable upper bound on the mean CD-KF posterior covariance under Poisson measurement arrivals coupled to auxiliary state dynamics, enabling gradient-based joint optimization of sensor rates and auxiliary inputs, plus a Wasserstein-based deterministic schedule for measurement times.
Problem
Scheduling and selecting among multiple sensors with irregular, discrete measurements in continuous-discrete Kalman filtering when measurement costs and accuracy are coupled to auxiliary state dynamics (energy, position, radiation damage, fouling).
Approach
Model sensor measurements as independent Poisson processes with sensor-specific rates; bound the mean posterior covariance (and concave/convex auxiliary state mean) via ODEs, yielding a continuously differentiable finite-horizon optimal control problem over rates and auxiliary inputs; then convert optimized rates into deterministic measurement times by optimal quantization under the Wasserstein-2
Key results
- In the robot GP-regression example, the Optimized method achieves mean covariance trace 1.90 versus 2.23 for Random and 2.61 for Greedy.
- Optimized mean energy is 21.5 versus -1.67 for Greedy and -17.4 for Random, and mean degradation 0.091 versus 0.193 for Greedy and 0.792 for Random.
- Deterministic quantization-based scheduling performs similarly to M-Optimized (trace 1.90 vs 1.92) without sampling multiple realizations.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| use deterministic quantization of the optimized measurement rates instead of greedy measurement selection | Robot example with radiation damage · Robot example with radiation damage | covariance trace mean: 2.60768 -> 1.90221 | Helped |
| use deterministic quantization of the optimized measurement rates instead of random constant-rate scheduling | Robot example with radiation damage · Robot example with radiation damage | covariance trace mean: 2.2275 -> 1.90221 | Helped |
| use deterministic quantization of the optimized measurement rates instead of sampled, lowest-cost schedules | Robot example with radiation damage · Robot example with radiation damage | covariance trace mean: 1.92461 -> 1.90221 | Helped |
| use deterministic quantization of the optimized measurement rates instead of greedy measurement selection | Robot example with radiation damage · Robot example with radiation damage | energy mean: -1.67343 -> 21.5435 | Helped |
| use deterministic quantization of the optimized measurement rates instead of random constant-rate scheduling | Robot example with radiation damage · Robot example with radiation damage | energy mean: -17.4417 -> 21.5435 | Helped |
| use deterministic quantization of the optimized measurement rates instead of sampled, lowest-cost schedules | Robot example with radiation damage · Robot example with radiation damage | energy mean: 22.3106 -> 21.5435 | Hurt |
| use deterministic quantization of the optimized measurement rates instead of greedy measurement selection | Robot example with radiation damage · Robot example with radiation damage | degradation mean: 0.193289 -> 0.0905529 | Helped |
| use deterministic quantization of the optimized measurement rates instead of random constant-rate scheduling | Robot example with radiation damage · Robot example with radiation damage | degradation mean: 0.792406 -> 0.0905529 | Helped |
| use deterministic quantization of the optimized measurement rates instead of sampled, lowest-cost schedules | Robot example with radiation damage · Robot example with radiation damage | degradation mean: 0.0845751 -> 0.0905529 | Hurt |
Limitations
- Covariance bound is only guaranteed along the mean auxiliary state; nonlinear or non-concave/convex auxiliary dynamics lack formal guarantees.
- Numerical direct collocation may not strictly satisfy constraints, and finer grids increase computational cost.
Experimental setup
Metrics: trace of posterior covariance energy level degradation
Baselines: Random (Poisson with constant rate) Greedy scheduling M-Optimized (sampled Poisson realizations)
Builds on: continuous-discrete Kalman filter Rauch-Tung-Striebel smoothing Matérn state-space GP representation
Closest papers by meaning
- Predictively-Oriented Kalman Filtering (NeurIPS 2026 · similarity 0.85)
- Implicit Maximum a Posteriori Filtering via Adaptive Optimization (ICLR 2024 · similarity 0.84)
- Max Entropy Moment Kalman Filter for Polynomial Systems with Arbitrary Noise (NeurIPS 2025 · similarity 0.83)
- TIDES: Implicit Time-Awareness in Selective State Space Models (NeurIPS 2026 · similarity 0.81)
- Christoffel-DPS: Optimal sensor placement in diffusion posterior sampling for arbitrary distributions (NeurIPS 2026 · similarity 0.81)
- Learning Dynamics under Environmental Constraints via Measurement-Induced Bundle Structures (ICML 2025 · similarity 0.81)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.