Research

Inertial Navigation Aided by Field Dynamics

This research aims at an under-explored aspect of underwater navigation, i.e. long-term, mid-depth navigation.  The mid-depth ocean, or the Bathypelagic zone, extends from a depth of 1,000 to 4,000 m below the ocean surface.  Albeit the crucial significance of persistently surveying and monitoring the Bathypelagic zone, long-term navigation within this depth range is extremely challenging due to the lack of natural or man-made navigation references.  To this end, I pioneered a flow-aided inertial navigation solution (see figure) that is fundamentally different from other existing localization solutions.  This navigation solution aims at improving the inertial navigation performance by utilizing local measurements of relative flow velocities and flow velocity predictions provided beforehand by ocean general circulation models (OGCM).  This method is implemented as a nonparametric, marginalized scholastic filter that is computationally suitable for compact AUVs.  I have demonstrated navigation accuracy [≤25% positioning uncertainty per distance traveled (UDT)] that is comparable to or better than state-of-the-art solutions while reducing the required navigational sensor grades (industrial or tactical grade) and extending the mission duration (tested for up to 25 hours).

References:

• Z. Song and K. Mohseni. Long-term inertial navigation aided by dynamics of flow field features. IEEE Journal of Oceanic Engineering, pp(99):1–15, 2017.
• Z. Song and K. Mohseni. Cooperative mid-depth navigation aided by ocean current prediction. In Proc. MTS/IEEE Oceans Conf., Anchorage, AK, USA, Sept 2017. [Student Poster Competition Third Prize; Poster]

---

Cooperative Localization

Without GPS coverage underwater, underwater robots often rely on expensive inertial navigation systems (INS) and acoustic devices in maintaining their navigation accuracy.  I am particularly interested in alternative solutions that utilize information exchange among collaborating AUVs, i.e. cooperative localization (CL).  Many existing approaches do not transform to underwater applications directly.  Besides the well-identified challenges such as large communication delay and the lack of natural global references, the presence of strong background flows often leads to stochastic robot networking topologies.  To tackle the challenges in tracking the correlation among robots’ state estimates resulted from information fusion, I proposed an extended Kalman filter (EKF) based information fusion approach.  This approach is fully distributed and robust towards uncertain networking topology changes due to the influence from background flows.  I applied this approach to a mother-daughter CL hierarchy that utilizes a small number of well-equipped mother AUVs as moving landmarks to maintain the CL accuracy of many daughter AUVs.  This method expands the range of AUVs while reducing the sensor and infrastructure costs for comparative navigation accuracy achievable by non-collaborative AUVs.

The flow-aided navigation scheme can also be extended to a multi-AUV navigation solution, dubbed flow-aided cooperative navigation (FACON), by allowing distributed information fusion among agents to achieve further improvement in navigation performance.  The figure above illustrates an exampling of FACON with covariance intersection.

References:

• • Z. Song and K. Mohseni. FACON: A flow-aided cooperative navigation scheme. In Proc. IEEE/RSJ Int. Conf. Intelligent Robots and Systems (IROS), Vancouver, BC, Canada, Sept 2017.

• Z. Song and K. Mohseni. Hierarchical underwater localization in dominating background flow fields. In Proc. IEEE/RSJ Int. Conf. Intelligent Robots and Systems (IROS), pages 3356–3361, Tokyo, Japan, Nov 2013.

---

Distributed Multi-Robot Swarming and Guidance

One of my recent research interests is the control and guidance of mobile robot swarms within strong background flows in adaptive and optimized fashions.  My collaborators and I consider the emergent robot swarm and the underlying geophysical flows as a single, integrated dynamical system.  We believe that modeling swarms as continuous fluids enables much convenience in (1) manipulating the macroscopic swarm dynamics in dynamic background flows, and (2) evaluating and predicting the performance of robot swarms within fluid environments.  These capabilities have been emphasized by several very recent research solicitations sponsored by DoD agencies.  We model robotic swarms using a numerical fluid simulation method named smoothed particle hydrodynamics (SPH).  This method considers each robot as a fluid parcel interacting with only adjacent neighbors.  All three essential properties possessed by well-behaved flocks, i.e. separation, cohesion and alignment, are naturally satisfied.  Flock guidance and obstacle avoidance are handled gracefully through the introduction of virtual attractors and repellers.  Through dimensional analysis we discovered that swarm compressibility and velocity consensus can be characterized and controlled by the Mach number and the Reynolds number, respectively, of the fluid emulated by the robot swarm.  We also demonstrated that (nearly) optimal guidance of SPH flocks within geophysical flows can be simplified to optimal path planning for a single agent.

References:

• • M. Silic, Z. Song and K. Mohseni. Anisotropic flocking control of distributed multi-agent systems using fluid abstraction. AIAA Information Systems-AIAA Infotech @ Aerospace, AIAA Paper-2018-2262, Kissimmee, FL, USA, Jan 2018.

• • Z. Song, D. Lipinski and K. Mohseni. Multi-vehicle cooperation and nearly fuel-optimal flock guidance in strong background flows. Ocean Engineering, 141:388–404, 2017.

• Z. Song and K. Mohseni. Anisotropic active Lagrangian particle swarm control in a meandering jet. In Proc. 54th IEEE Conf. Decision and Control (CDC), pages 240–245, Osaka, Japan, Dec 2015.

---

Marine Robot Prototyping

References:

• Z. Song, C. Mazzola, E. Schwartz, J. Finlaw, R. Chen, M. Krieg and K. Mohseni. A compact autonomous underwater vehicle with cephalopod-inspired propulsion. Marine Technology Society Journal, 50(5):88–101, 2016.

---

Optimal Motion Planning with Plenoptic Imaging

This project focuses on three major components of a visual guidance system for automated underwater vehicle docking: target acquisition with plenoptic imaging, optimal motion planning, and nonlinear trajectory tracking control.  With the camera sensor data of each photographic exposure, depth maps of the scene are generated through light-field tomographic reconstruction to provide relative position update of the docking station.  A variational approach is adopted to optimize the resulting depth maps for improved guidance accuracy.  Assisted by color detection, target docking positions for the vehicle are generated based on the knowledge of the docking structure, the depth maps, and the extrinsic camera calibration matrix.  A nonlinear optimal trajectory generator is then used to calculate the best docking trajectory by solving the nonlinear programing problem that simultaneously minimizes the mission time and actuation energy.  Asymptotic trajectory tracking performance is achieved through continuous, feedback control.  Uniform, time-invariant background flows are considered in optimal trajectory planning and trajectory tracking control design to generalize the proposed docking control method for missions with ocean currents or docking structures in regulated motions.

References:

• Z. Song and K. Mohseni. Automated AUV docking control with light-field imaging. In Proc. MTS/IEEE Oceans Conf., Anchorage, AK, USA, Sept 2017.

---

Bio-inspired Contrast-based Underwater Visual Guidance

In this project, an obstacle avoidance strategy using monocular gray-scale robotic vision for turbid water environments is be investigated.  Biologically inspired by the unique vision system of the cubozoan, or box jellyfish, the proposed obstacle avoidance techniques were designed to be as computationally inexpensive as possible for implementation in a compact autonomous underwater vehicle with on-board processing capabilities.  The sharp contrast reduction in turbid waters between obstacles and the surrounding environment is leveraged as a semi-reliable measure of relative distance between obstacles to form an evasion response based on obstacle priority.  This contrast reduction model can be applied to both underwater and aerial vehicles depending on environmental turbidity and the relative distance between obstacles.  In order to test this bio-inspired approach, the proposed obstacle avoidance algorithm is implemented on a simple, low-power digital signal processor.  It is shown that using contrast as a sole depth cue in turbid underwater environments is suitable for the detection of large, stationary obstacles.

References:

• Z. Song, E. Schwartz and K. Mohseni. Bioinspired visual guidance in turbid underwater environment. In Proc. IEEE Sensors Conf., Glasgow, Scotland, UK, Oct 2017. [Poster]

---

Underwater Visible Light Communication

This project focuses on the design and implementation of a low-power, optical communication system for compact underwater autonomous vehicles for short-range, high-bandwidth data transmission, with an emphasis on identifying optimal hardware configurations for maximizing transmission distance or link misalignment tolerance.  Theoretical power and misalignment models are investigated for a single static, light-emitting diode as transmission source, with optical lenses on both transmitting and receiving ends of the system.  The maximum misalignment tolerance, while performing digital data communication, is also being studied.  We showed that the trade-off between transmission distance and misalignment tolerance is largely determined by the focus level of the receiver lens, which can be controlled by the position of the receiver photodiode on the optical axis.

References:

• Z. Song, E. Schwartz and K. Mohseni. A low-power optical communication modem for compact autonomous underwater vehicles. In Proc. IEEE Sensors Conf., Glasgow, Scotland, UK, Oct 2017. [Best Student Paper Award Finalist; Poster]