old | Shreyas Ramakrishna

Autonomous Driving Testbed

Mon, 01 Jan 0001 00:00:00 +0000

To implement, validate, and test our research products, we have mainly used the CARLA simulator and the DeepNNCar. While CARLA is a well-known open-source urban driving simulator, DeepNNCar is a low-cost research testbed that was designed in the Smart and Resilient Computing for Physical Environments Lab (SCOPE). DeepNNCar is built upon the chassis of Traxxas Slash 2WD 1/10 Scale RC car and is mounted with a USB forward-looking camera, IR- optocoupler, and a 2D LIDAR. The speed and steer for the robot are controlled using pulse-width modulation (PWM), by varying the duty cycle. [Recommended Reading] [Web Content]

Demonstration

A demonstration of the DeepNNCar operating on an indoor track. The car performs end-to-end driving using a modified NVIDIA DAVE-II convolutional neural network. The network is trained to drive within the tracks while achieving high speeds. You can learn more about the platform from our GitHub

Design-Time Assurance Case Construction

Mon, 01 Jan 0001 00:00:00 +0000

As assurance cases have become an integral component for safety-certification in various CPS domains, including automotive, aviation, military, and medical devices. Despite the strict requirements, current practices still rely on manual methods that are brittle, do not have a systematic approach or thorough consideration of sound arguments. In addition, stringent certification requirements and ever-increasing system complexity make ad-hoc, manual assurance case generation inefficient, time consuming, and expensive. In this area, we improve the current state of practice by introducing a structured assurance case generation (ACG) method which uses system design artifacts, accumulated evidence, and developer expertise to construct an assurance case and evaluate it in an automated manner. Besides, we also focus on designing a tool called ALC to automate the entire design and testing of learning-enabled CPSs. Recommended Reading

Out-of-Distribution Detection and Feature Identification

Mon, 01 Jan 0001 00:00:00 +0000

Machine Learning components have shown remarkable performance in several perception and control tasks like NVIDIA’s DAVE-II self-driving car. However, incidents like TESLA’s self-driving accident and UBER’s autonomous car crash have shown these components to be susceptible to Out-of-distribution (OOD) data. Besides, the black-box nature of these components makes it difficult to test and verify them. To address this, we have used a generative model called B-Variational Autoencoder (B-VAE) to detect the OOD data and identify the factor responsible for the OOD data problem. For example, if an autonomous vehicle is trained on images from day scenes, but if it encounters images from evening scenes, then the vehicle’s performance on these images will be erroneous. For the vehicle’s safety, it is required to detect that the operating scene has changed from that of training and the time-of-day factor is responsible for the problem. Recommended Reading

Demonstration

The precipitation increases to a high value at t = 13 seconds in the video. Images of high rain were not included in training the Machine Learning controller driving the AV. So, we used the B-VAE detector to identify OOD images in CARLA simulation. As the precipitation value increases, the detector martingale and the martingale of the precipitation reasoner increases (Martingale used for performing time-series detection. You can find out more from the recommended reading material) You can find more videos and the implementation from our GitHub.

Proactive Decision Making for Autonomous Systems

Mon, 01 Jan 0001 00:00:00 +0000

Adaptive decision-making for mitigation is a decisive step of my dynamic assurance framework. Once a problem has been detected, and the system risk is assessed, this information needs to be proactively used for selecting a suitable control action that mitigates the system’s risk. Simplex Architecture has been widely utilized as a mitigation strategy in Cyber-Physical Systems. This architecture integrates the system to be assured with a safe controller and provides a decision logic to switch between the decisions of these controllers. However, the decision logic is always designed “offline” and used at runtime for control action selection. Such an offline logic cannot optimally balance the safety vs. mission-critical information (e.g., performance) of the system employing the architecture. I have been working towards a “proactive” and “adaptive” decision logic to address this limitation. I have used reinforcement learning in my previous work, and I plan to use Monte Carlo Tree Search (MCTS) to learn an online decision logic at runtime. Recommended Reading