About

Gabriel Hugh Elkaim is a professor at UC Santa Cruz, affiliated with the Baskin School of Engineering in the Department of Electrical and Computer Engineering. His research focuses on autonomous and embedded systems, with particular interests in sensor fusion, guidance, navigation, and control (GNC), system identification, robust software architectures for real-time reactive systems, and robotics, including unmanned autonomous vehicles (UAVs) and cooperative control. He has a background in Mechanical and Aerospace Engineering from Princeton University, and holds both a Master’s and Ph.D. from Stanford University, where he built an autonomous catamaran for his thesis. Elkaim has extensive international experience, having worked in oilfield services with Schlumberger in Algeria and Nigeria, and studied Aeronautics at the Technion in Israel. Since joining UC Santa Cruz in 2003, he has established the Autonomous Systems Lab and has been involved in teaching courses such as Mechatronics, Linear Dynamical Systems, Feedback Control Applications, and UAV Modeling and Control. His professional interests include robotics, autonomous vehicles, and control systems.

Research topics

• Computer Science
• Artificial Intelligence
• Mathematics
• Simulation
• Computer vision

Selected publications

• GNSS-Assisted System Identification of Autonomous Ground Vehicle Model and Sensor Parameters

  2023-04-24 · 3 citations

  articleSenior author

  Vehicle autonomy requires accurate modeling of both kinematic and dynamic parameters of a given craft. While many of these parameters may be measured directly or determined through experimentation, it is often desirable to measure and update them in the field, that is, while the vehicle is operating. The challenge with online system identification, however, is the lack of an absolute reference to determine the true value of a given variable of interest. When used appropriately, global navigation satellite system (GNSS) sensors are able to provide such a reference for these variables. Furthermore, these sensors can operate at a high enough frequency to provide near real-time feedback to the vehicle. A related challenge for autonomous vehicles is sensor calibration. Many inexpensive inertial measurement units (IMU) are subject to drift or error due to changes in the environment. For these challenges the GNSS sensor may also provide a useful reference to determine sensor drift as well as a means for recalibration. In this work we present methods for addressing several challenges relevant to autonomous ground vehicles using a GNSS sensor as an absolute reference. We have built a small scale autonomous ground vehicle (AGV) as a test platform equipped with onboard odometry, an IMU and a commercial GNSS sensor. We initially demonstrate methods using GNSS data to determine parameters of the dynamic and kinematic models. We determine the coefficients of a frequency domain model of the AGV drivetrain using on a simple DC motor and an autoregresson with external inputs (ARX) system identification technique. We extend this model to determine the longitudinal transfer function of the system using step function tests and GNSS-derived velocity measurements. We also demonstrate methods to determine the AGV kinematic model parameters: turning radius, wheelbase, effective tire radius of the AGV, and steering mechanism parameters. We compare lab measurements and against results obtained with the GNSS sensor using least-squares fitting methods. Using the GNSS-derived kinematic parameters we determine the best estimate of the AGV odometry model in comparison to the absolute position and heading as determined by the GNSS sensor. Lastly, we demonstrate methods to calibrate two onboard sensors. The first method collects magnetometer readings in the local tangent plane and computes calibration factors using an iterative least squares fit of the data to a unit circle. The second method calibrates the yaw rate as measured by the <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$z$</tex> - axis gyroscope sensor using the GNSS-derived rate of change of heading angle.

  PublisherDOI

• A Deep Reinforcement Learning Approach for Long-term Short-term Planning on Frenet Frame

  2022 IEEE 18th International Conference on Automation Science and Engineering (CASE) · 2021 · 18 citations

  Senior authorCorresponding
  • Computer Science
  • Artificial Intelligence
  • Computer Science

  Tactical decision-making and strategic motion planning for autonomous highway driving are challenging due to predicting other road users' behaviors, diversity of environments, and complexity of the traffic interactions. This paper presents a novel end-to-end continuous deep reinforcement learning approach towards autonomous cars' decision-making and motion planning. For the first time, we define both states and action spaces on the Frenet space to make the driving behavior less variant to the road curvatures than the surrounding actors' dynamics and traffic interactions. The agent receives time-series data of past trajectories of the surrounding vehicles and applies convolutional neural networks along the time channels to extract features in the backbone. The algorithm generates continuous spatiotemporal trajectories on the Frenet frame for the feedback controller to track. Extensive high-fidelity highway simulations on CARLA show the superiority of the presented approach compared with commonly used baselines and discrete reinforcement learning on various traffic scenarios. Furthermore, the proposed method's advantage is confirmed with a more comprehensive performance evaluation against 1000 randomly generated test scenarios. Code: https://github.com/MajidMoghadam2006/RL-frenet-trajectory-planning-in-CARLA

  PublisherDOI

• An Autonomous Driving Framework for Long-Term Decision-Making and Short-Term Trajectory Planning on Frenet Space

  2021-08-23 · 3 citations

  preprintOpen accessSenior author

  In this paper, we present a hierarchical framework for decision-making and planning on highway driving tasks. We utilized intelligent driving models (IDM and MOBIL) to generate long-term decisions based on the traffic situation flowing around the ego. The decisions both maximize ego performance while respecting other vehicles&#x0027; objectives. Short-term trajectory optimization is performed on the Frenet space to make the calculations invariant to the road&#x0027;s three-dimensional curvatures. A novel obstacle avoidance approach is introduced on the Frenet frame for the moving obstacles. The optimization explores the driving corridors to generate spatiotemporal polynomial trajectories to navigate through the traffic safely and obey the BP commands. The framework also introduces a heuristic supervisor that identifies unexpected situations and recalculates each module in case of a potential emergency. Experiments in CARLA simulation have shown the potential and the scalability of the framework in implementing various driving styles that match human behavior.

  PublisherOA PDFDOI

• An Autonomous Driving Framework for Long-term Decision-making and\n Short-term Trajectory Planning on Frenet Space

  arXiv (Cornell University) · 2020-11-25 · 1 citations

  preprintOpen accessSenior author

  In this paper, we present a hierarchical framework for decision-making and\nplanning on highway driving tasks. We utilized intelligent driving models (IDM\nand MOBIL) to generate long-term decisions based on the traffic situation\nflowing around the ego. The decisions both maximize ego performance while\nrespecting other vehicles' objectives. Short-term trajectory optimization is\nperformed on the Frenet space to make the calculations invariant to the road's\nthree-dimensional curvatures. A novel obstacle avoidance approach is introduced\non the Frenet frame for the moving obstacles. The optimization explores the\ndriving corridors to generate spatiotemporal polynomial trajectories to\nnavigate through the traffic safely and obey the BP commands. The framework\nalso introduces a heuristic supervisor that identifies unexpected situations\nand recalculates each module in case of a potential emergency. Experiments in\nCARLA simulation have shown the potential and the scalability of the framework\nin implementing various driving styles that match human behavior.\n

  PublisherOA PDFDOI

• Batch Misalignment Calibration of Multiple Three-Axis Sensors

  Springer proceedings in advanced robotics · 2020-01-01

  book-chapter1st authorCorresponding
  PublisherDOI

• An End-to-end Deep Reinforcement Learning Approach for the Long-term Short-term Planning on the Frenet Space

  arXiv (Cornell University) · 2020 · 6 citations

  Senior authorCorresponding
  • Computer Science
  • Artificial Intelligence
  • Computer Science

  Tactical decision making and strategic motion planning for autonomous highway driving are challenging due to the complication of predicting other road users' behaviors, diversity of environments, and complexity of the traffic interactions. This paper presents a novel end-to-end continuous deep reinforcement learning approach towards autonomous cars' decision-making and motion planning. For the first time, we define both states and action spaces on the Frenet space to make the driving behavior less variant to the road curvatures than the surrounding actors' dynamics and traffic interactions. The agent receives time-series data of past trajectories of the surrounding vehicles and applies convolutional neural networks along the time channels to extract features in the backbone. The algorithm generates continuous spatiotemporal trajectories on the Frenet frame for the feedback controller to track. Extensive high-fidelity highway simulations on CARLA show the superiority of the presented approach compared with commonly used baselines and discrete reinforcement learning on various traffic scenarios. Furthermore, the proposed method's advantage is confirmed with a more comprehensive performance evaluation against 1000 randomly generated test scenarios.

  PublisherOA PDFDOI

• Uncertainty Suppression Methods for the Exploration of Sparsely Sampled Fields

  Proceedings of the Satellite Division's International Technical Meeting (Online)/Proceedings of the Satellite Division's International Technical Meeting (CD-ROM) · 2019-10-11 · 1 citations

  articleSenior author

  The exploration of unknown areas has been accelerated by the development of consumer grade UAVs. Sensors and sensing systems have also become commodity items that researchers and developers can use to gain an understanding of the composition and concentrations of nutrients, moisture, and pollutants in soil, water, and air. A large unknown field of interest (a target field) can be explored to a high degree of accuracy if the entire field could be scanned using any of the various autonomous exploration system. This may not be possible if the field is very large and the exploration vehicle has limited exploration time due to terrain restrictions, battery/fuel limitations, and/or sensing capabilities. If the quantity of interest being predicted in a target field exhibits a degree of spatial autocorrelation, then estimates can be made at all points in the field based on observations taken at a few selected points. This paper explores where an exploration vehicle should travel to collect samples that create minimum variance estimates of all points in the field. It does this by determining the spatial autocorrelation of the estimated quantity, and uses this information to determine the next portion of the path that minimizes the sum of the variances of every point in the field. The Kriging Method, a Best Linear Unbiased Predictor, generates a prediction, and variance of that prediction, of a single point in a target field given a set of observed points. Averaging the estimation variances across all points in the field provides an objective function to guide path planning. The goal of each of the planners introduced is to assist in the discovery of a field’s features with an adjustable trade-off between arc length of the total path traveled and confidence of field prediction. Each of the path planners introduced attempts to reduce Kriging prediction variance by steering an exploration vehicle through a target field in a fashion that is predicted to reduce overall field uncertainty.

  PublisherDOI

• A Hierarchical Architecture for Sequential Decision-Making in Autonomous Driving using Deep Reinforcement Learning

  arXiv (Cornell University) · 2019-06-20 · 14 citations

  preprintOpen accessSenior author

  Tactical decision making is a critical feature for advanced driving systems, that incorporates several challenges such as complexity of the uncertain environment and reliability of the autonomous system. In this work, we develop a multi-modal architecture that includes the environmental modeling of ego surrounding and train a deep reinforcement learning (DRL) agent that yields consistent performance in stochastic highway driving scenarios. To this end, we feed the occupancy grid of the ego surrounding into the DRL agent and obtain the high-level sequential commands (i.e. lane change) to send them to lower-level controllers. We will show that dividing the autonomous driving problem into a multi-layer control architecture enables us to leverage the AI power to solve each layer separately and achieve an admissible reliability score. Comparing with end-to-end approaches, this architecture enables us to end up with a more reliable system which can be implemented in actual self-driving cars.

  PublisherOA PDFDOI

• Filtering and Sensor Augmentation for GPS Measurement Reduction in Wildlife Tags

  Proceedings of the Satellite Division's International Technical Meeting (Online)/Proceedings of the Satellite Division's International Technical Meeting (CD-ROM) · 2019-10-11 · 1 citations

  articleSenior author

  Animal-mounted GPS-based location tracking has become a core tool for wildlife ecologists. However, the lifetimes of animal mounted devices (or “tags”) are typically limited by battery life, and GPS tracking occupies a large portion of their energy budgets. In this paper, we propose and test several Kalman filter-based algorithms that reduce the GPS duty cycle while still maintaining a threshold of tracking accuracy. These algorithms leverage low-power accelerometry measurements to estimate the uncertainty in the tag’s location, then schedule GPS measurements to suppress that uncertainty. We show that these strategies can reduce average GPS uptime, though at the cost of fidelity in some cases.

  PublisherDOI

• Attitude Estimation for Small, Low-Cost UAVs

  2016-08-11 · 3 citations

  article1st authorCorresponding
  PublisherDOI

Recent grants

• MRI: Development of an Autonomous Robotic Vehicle Instrument (ARVIN)

  NSF · $376k · 2005–2012

Frequent coauthors

• Renwick E. Curry

  University of California, Santa Cruz

  9 shared

• Ji-wung Choi

  Hyundai Mobis (South Korea)

  5 shared

• Majid Moghadam

  Birjand University of Medical Sciences

  5 shared

• Mariano Lizarraga

  Amazon (United States)

  4 shared

• C.O. Lee Boyce

  4 shared

• Engin Tekin

  University of California, Santa Cruz

  2 shared

• David Ilstrup

  2 shared

• Fidelis Adhika Pradipta Lie

  University of Minnesota System

  2 shared

Labs

• The Autonomous Systems LabPI