Research map / Probabilistic and scientific modeling / Molecules and proteins
Power-Aware Inverse-Search Machine Learning for Low Resource Multi-Objective Unmanned Underwater Vehicle Control (Student Abstract)
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.
Adds a forward LSTM power model to a search-based inverse model for flapping-fin UUV control, optimizing thrust, kinematic smoothness, and power via weight combinations, and benchmarks it on a Raspberry Pi 4B.
Problem
Flapping-fin UUV control needs real-time inverse search that maps thrust requests to gaits while balancing thrust, smooth transitions, and power consumption under limited compute and battery.
Approach
Extend a previous two-objective (thrust, kinematic smoothness) inverse search model with an LSTM forward power predictor so the search also minimizes power, and sweep objective weights and fin materials to define movement modes.
Key results
- Weight (0.8, 0.2, 0) reduces average power loss by about 3 W for only about 0.1 N added thrust loss relative to thrust-only weighting.
- Thrust loss is about 0.2 N when thrust weight is 0.8 or higher and rises to about 0.5 N below 0.8.
- The PDMS 1:10 fin has lower average power loss (about 1.2 W) than the rigid fin (about 1.5 W).
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Use a thrust weight below 0.8 in the three-objective weight sweep (ablation) | Power-aware inverse search · All tested fin designs, weight sweep | thrust loss: around 0.2 N -> around 0.5 N average loss | Hurt |
| Set weights around (wT, wP, wK) = (0.8, 0.2, 0) (ablation) | Power-aware inverse search · All tested fin designs, weight sweep | power loss: by a factor of 3 W | Helped |
| Set weights around (wT, wP, wK) = (0.8, 0.2, 0) (ablation) | Power-aware inverse search · All tested fin designs, weight sweep | thrust loss: 0.1 N trade off in thrust | Hurt |
| Set weights to (wT, wP, wK) = (1, 0, 0) (ablation) | Power-aware inverse search · All tested fin designs, weight sweep | thrust loss | Helped |
| Use flexible PDMS 1:10 or PDMS 1:20 instead of the rigid fin | Power-aware inverse search · Weight combinations | thrust loss: averaging around 0.5N | Mixed |
| Use PDMS 1:10 instead of either other fin design | Power-aware inverse search · Weight combinations | power loss: around 1.2W compared to 1.5W | Helped |
Surprising findings
- Power loss and kinematic smoothness loss behave similarly across weight combinations, since switching between very different movements raises power consumption.
Limitations
- Final time-performance benchmarking and physical-system validation are left for future work; only simulated movements are tested.
- Power model adds computational power draw and time, and the timing claim is not fully verified in the paper.
Experimental setup
Datasets: 864 unique flapping-fin gaits with thrust and power measurements for rigid, PDMS 1:10, and PDMS 1:20 fins
Benchmarks: simulated thrust requests across weight combinations Raspberry Pi 4B timing benchmark
Metrics: thrust loss power loss kinematic loss computation time
Baselines: two-constraint inverse model (prior work)
Builds on: Lee et al. 2023 inverse model for UUVs LSTM surrogate models generalized pattern search
Closest papers by meaning
- Reducing AUV Energy Consumption Through Dynamic Sensor Directions Switching via Deep Reinforcement Learning (AAAI 2025 · similarity 0.84)
- A deep inverse-mapping model for a flapping robotic wing (ICLR 2025 · similarity 0.84)
- UACOF: A USV-AUV Collaboration Framework for Underwater Tasks Under Extreme Sea Conditions (Student Abstract) (AAAI 2025 · similarity 0.80)
- Markovian Dynamics Enforcer: Feasibility Preserving Correction on Learned Dyanmics Manifolds (NeurIPS 2026 · similarity 0.80)
- Predictive Inverse Dynamics Models are Scalable Learners for Robotic Manipulation (ICLR 2025 · similarity 0.80)
- Zero-Shot Transfer of Neural ODEs (NeurIPS 2024 · similarity 0.79)
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.