Reliability-Adaptive Multimodal Relative Pose Tracking for Underwater Leader-Follower Robots
Reliability-Adaptive Multimodal Relative Pose Tracking for Underwater Leader-Follower Robots
This paper presents a reliability-adaptive multimodal perception framework for six-degree-of-freedom relative-pose tracking between two moving underwater robots. The framework integrates visual, acoustic-ranging, and hydrodynamic measurements while adaptively weighting modality contributions based on time-varying reliability. Extensive experiments using a custom-built AUV platform across 648 trials show an average translational error of 3.8 cm and orientation error of 2.4 degrees. Onboard NVIDIA Jetson AGX Orin, the fusion network achieves 17.62 ms inference latency, offering a 2.69-fold speedup over a Transformer-based baseline. Closed-loop tests confirm stable formation maintenance under computational constraints, and a multimodal dataset from all trials will be released publicly.
The study demonstrates a reliability-adaptive multimodal framework achieving 3.8 cm translational error and 2.4 degree orientation error on an AUV platform with 17.62 ms inference latency, validated across 648 pool trials under varying illumination and trajectory conditions.
Source: arXiv Robotics — research abstracts · Read original article ↗
Loading article text…
What the source reports
Publisher-reported claims, with original evidence. These results have not been independently verified by RoboSignal.
Reported numbers
trials
648
Reported trials
View original evidence
stic-ranging, and hydrodynamic measurements while accounting for their distinct physical characteristics and time-varying reliability. It adaptively weights modality contributions and models temporal dependencies with linear computational complexity, enabling efficient pose estimation on embedded hardware. Extensive experiments were conducted using a custom-built AUV platform, with 648 pool trials spanning two leader
Open source S3translational error
3.8
View original evidence
vehicle types, three illumination conditions, and nine trajectory patterns. Evaluation yields an average translational error of 3.8 cm and an average orientation error of 2.4{\deg}. On an onboard NVIDIA Jetson AGX Orin, the fusion network achieves an inference latency of 17.62 ms, providing a 2.69-fold speedup over the evaluated Transformer-based fusion baseline. Closed-loop experiments further demonstrate stable fo
Open source S4orientation error
2.4
View original evidence
vehicle types, three illumination conditions, and nine trajectory patterns. Evaluation yields an average translational error of 3.8 cm and an average orientation error of 2.4{\deg}. On an onboard NVIDIA Jetson AGX Orin, the fusion network achieves an inference latency of 17.62 ms, providing a 2.69-fold speedup over the evaluated Transformer-based fusion baseline. Closed-loop experiments further demonstrate stable fo
Open source S4inference latency
17.62
View original evidence
vehicle types, three illumination conditions, and nine trajectory patterns. Evaluation yields an average translational error of 3.8 cm and an average orientation error of 2.4{\deg}. On an onboard NVIDIA Jetson AGX Orin, the fusion network achieves an inference latency of 17.62 ms, providing a 2.69-fold speedup over the evaluated Transformer-based fusion baseline. Closed-loop experiments further demonstrate stable fo
Open source S4speedup factor
2.69
View original evidence
vehicle types, three illumination conditions, and nine trajectory patterns. Evaluation yields an average translational error of 3.8 cm and an average orientation error of 2.4{\deg}. On an onboard NVIDIA Jetson AGX Orin, the fusion network achieves an inference latency of 17.62 ms, providing a 2.69-fold speedup over the evaluated Transformer-based fusion baseline. Closed-loop experiments further demonstrate stable fo
Open source S4leader vehicle types
2
View original evidence
stic-ranging, and hydrodynamic measurements while accounting for their distinct physical characteristics and time-varying reliability. It adaptively weights modality contributions and models temporal dependencies with linear computational complexity, enabling efficient pose estimation on embedded hardware. Extensive experiments were conducted using a custom-built AUV platform, with 648 pool trials spanning two leader
Open source S3
- dataset: Reported pending Open source S5
- code: not_reported
- weights: not_reported
- dataset: not_reported
- paper: not_reported
Source excerpts and review record
Automatically extracted; no manual editorial approval recorded.
stic-ranging, and hydrodynamic measurements while accounting for their distinct physical characteristics and time-varying reliability. It adaptively weights modality contributions and models temporal dependencies with linear computational complexity, enabling efficient pose estimation on embedded hardware. Extensive experiments were conducted using a custom-built AUV platform, with 648 pool trials spanning two leader
Open source S3
rmation maintenance with a moving leader under onboard computational constraints. A multimodal dataset containing measurements from all 648 trials will be made publicly available.
Open source S5
vehicle types, three illumination conditions, and nine trajectory patterns. Evaluation yields an average translational error of 3.8 cm and an average orientation error of 2.4{\deg}. On an onboard NVIDIA Jetson AGX Orin, the fusion network achieves an inference latency of 17.62 ms, providing a 2.69-fold speedup over the evaluated Transformer-based fusion baseline. Closed-loop experiments further demonstrate stable fo
Open source S4
Source：arXiv Robotics — research abstracts · arxiv.org