Autonomous Vehicle Laboratory - Contextual Robotics Institute


The Autonomous Vehicle Laboratory is a research team at UC San Diego led by the Contextual Robotics Director, Dr. Henrik. I. Christensen. The group's research focus is to explore and develop robust self-driving car systems and architectures. Under Dr. Christensen's supervision, David Paz has explored the capabilities and limitations of today's self-driving technology: localization and mapping using LiDAR technology, vision and LiDAR based obstacle detection and fusion, low-level controls, and planning.

The team is currently working on addressing some of the draw-backs of using LiDAR sensors, dense point cloud maps, and HD maps. Due to the constant maintenance required and the high costs of using LiDAR sensors, scalability is hindered.

With safety measures in place, the UC San Diego campus is also serving as a testing ground for self-driving cars. The two development platforms have been used to delivery mail autonomously to different locations on campus and last-mile transporation for campus visitors, students and patients. https://youtu.be/

Testing and early development (Fall 2018)

Mail Delivery Deployment (Summer 2019)

UC San Diego RoboCar


The UC San Diego RoboCar is an ongoing project that is being used to explore and learn more about autonomous vehicle technologies. In this initial phase, our team used steering, throttle and a camera as inputs to train a neural network model. Using low-cost components (Raspberry PI, camera, and PWM) , we were able to achieve autonomy after training the vehicle on a track for fifteen laps. The robocar's behavior was non-deterministic and displayed self-correcting capabilities when wrong turns were made. Future work involves using ROS and the NVIDIA Jetson TX2 to develop a model that is capable of navigating autonomously after creating a 3D map of the area.

Optimizing Two-Dimensional Convolution Accelerators for Area, Energy, and Flexibility


Two-dimensional convolution calculations have direct applications in image processing, filtering and pattern detection. Highly optimized hardware accelerators are capable of increasing performance on very specific computational tasks by large factors over sequential software applications. These accelerators are often optimized for instructions per clock cycle. However, one important aspect of hardware accelerator design that is often overlooked relates to application flexibility. For instance, Convolution Neural Network (CNN) accelerators are designed with a predefined number of accelerator cores and application specific tasks that may not be modified dynamically and can be expensive. This study aims to develop energy efficient and flexible two-dimensional convolutional accelerators ideal for IoT and smaller devices to provide significant performance gains over sequential computations and flexibility over application-specific accelerators such as CNN accelerators. Bluespec System Verilog (BSV) has been used to develop the accelerator by implementing pipelined Full Binary Tree structures. These structures have been fully incorporated into a three-stage pipelined RISC-V processor, and they are capable of interacting directly with memory and exploit temporal locality to maximize performance with minimal energy and area cost, while providing users with fine-grain control of their kernel and matrices.

Abstract: PDF
MSRP Poster (1-D and 2-D Accelerators): PDF

Performance Analysis of Applications using Singularity Container on SDSC Comet


This study aims to analyze the properties and HPC performance implications of Singularity to determine if container benefits amortize the overhead cost. Multiple nodes on the Comet supercomputer at the San Diego Supercomputer Center were used to analyze and compare (between Singularity and non-Singularity) performance of resource intensive applications and benchmark codes such as NEURON and Intel MPI Benchmarks (IMB). NEURON software was used to simulate a complex neural network model, and the OSU and IMB benchmarks were used to calculate the latency and communication efficiency for MPI. The containerized runtimes were directly compared with the corresponding non-containerized runtime of jobs to analyze the performance of each method. For future work, we plan to explore other technologies such as Shifter and study their performance on HPC.

ACM Publication: PDF
PEARC17 Poster: PDF

A Portable and Reusable Lead Detecting Device


In many developing countries, daily consumption of water from polluted sources is the cause of thousands of deaths. Although natural sources of water contain certain elements which are safe and essential for the human body, an estimated of 143,000 deaths were the result of lead exposure worldwide. In particular, exposure to high concentrations of lead can lead to brain and kidney damage. Efficient methods for the detection of lead and other harmful heavy metals have been developed; however, these methods are expensive or non-portable. This study aims to develop a low cost, portable electronic device that is capable of detecting high concentration of lead in water and provide real time updates to the user in order to prevent ingestion. Unlike other approaches, pairing graphene with Modified Multi-Walled Carbon Paste Nanotubes (MWCNT) allows us to effectively detect changes in lead concentrations due to their highly accurate and lead-selective properties. This process requires a potentiometric approach; the resistive element embedded in the sensor--composed of MWCNT--indicates an increase of lead ion concentration when the electric potential drops. The process of this detection is lead selective and can be mathematically calculated by using Ohm’s law to display and notify the user. With the use of UHF RFID technology(ultra-high frequency radio frequency identification) and NFC (near field communication), the signal from the sensor is detected and sent to a device to be interpreted.

Learn more
Abstract: PDF