Revolutionizing Goods Delivery with Nuro's Autonomy Stack

Table of Contents

1. Introduction
2. Our Autonomy Stack
   • Mapping and Localization
   • Perception: Understanding the Dynamic Environment
   • Prediction and Planning: Interacting with Other Traffic Participants
   • Remote Operations: Leveraging Human Intelligence
3. Designing Our Custom Vehicle
   • Full 360-Degree Sensor Coverage
   • Maximizing Safety in Challenging Scenarios
4. Validation Process for Safety on Public Roads
   • Hyper-Local Deployments and Tailored Requirements
   • Verification Methods
5. Scaling Autonomy for Mass Deployments
   • Scaling ML for Rapid Progress
   • Building a Holistic Validation Process
6. Building a Scalable Business
   • Operating and Maintaining the Fleet Remotely
   • Seamless Integration with Retailer Order Systems
   • Customer Experience: Ordering, Tracking, and Delivery
7. Ensuring Security
   • Trusted Vehicle Platform and Teleoperation System
   • Protecting Partner Data and User Information
   • Emulating Attacks to Verify Security
8. Conclusion

Our Autonomy Stack

Our autonomy stack is built around our zero occupant vehicle, and in this section, we will dive deep into the different components of our technology that make it unique and high-performing.

Mapping and Localization

In order to navigate autonomously, our vehicles rely on highly detailed maps that we build in-house. We use a combination of data collected by our operators driving selected routes and machine learning algorithms to Create maps that contain accurate information about lanes, traffic lights, crosswalks, road signs, and more. This approach allows us to have hyperlocal deployments with maps specifically tailored to our delivery areas, ensuring precision and efficiency.

Perception: Understanding the Dynamic Environment

Our perception system is designed to interpret the dynamic environment around the vehicle. By fusing information from various sensors, including cameras, lidar, and radar, we can accurately detect and track objects such as pedestrians, cars, and cyclists. Our perception stack is built on an ML-first approach, with robust fallbacks to ensure safety and reliability. We prioritize safety over comfort, taking cautious actions when there is uncertainty about the nature of an object or an interaction.

Prediction and Planning: Interacting with Other Traffic Participants

The prediction and planning module is responsible for interacting with other traffic participants and making decisions about the vehicle's path. Using machine learning models and analysis of the intents of other agents, we predict their future behavior and plan our own actions accordingly. Our goods-first approach allows us to be extra cautious in neighborhood driving, accommodating the presence of aggressive agents and ensuring the safety of our delivery operations.

Remote Operations: Leveraging Human Intelligence

In situations where the autonomy system encounters complex or unpredictable scenarios, we leverage remote operations to involve human intelligence. Our operators can provide different levels of assistance, from stopping the vehicle in hazardous situations to taking full control when necessary. Being passengerless, our vehicles can safely come to a stop and wait for a remote operator without impacting customer experience. The ability to remotely operate and assist our vehicles enhances safety and allows us to handle edge cases effectively.

Designing Our Custom Vehicle

To support our autonomy stack and ensure maximum safety and performance, we have designed a custom vehicle that meets our specific requirements.

Full 360-Degree Sensor Coverage

Our custom vehicle is equipped with a comprehensive suite of sensors, providing full 360-degree coverage for both cameras and lidar. This sensor configuration allows us to accurately detect and track Relevant objects in the vehicle's surroundings, regardless of the environmental variables or roadway configuration. We also strategically place thermal cameras in the front to enhance safety for pedestrians, cyclists, and animals.

Maximizing Safety in Challenging Scenarios

The design of our custom vehicle takes into consideration the challenges posed by various scenarios, such as hard intersections, lane splitting, and interactions with pedestrians and cyclists. By incorporating redundancy and leveraging the strengths of different sensors, we ensure reliable detections even in adverse conditions like rain or low light. Our smaller footprint and fully redundant sensor stack enable our vehicles to navigate narrow roads and neighborhood environments safely.

Validation Process for Safety on Public Roads

Safety is our utmost priority, and we have developed a rigorous validation process to ensure our technology is ready for deployment on public roads.

Hyper-Local Deployments and Tailored Requirements

Before any deployment, we define requirements specific to the intended operating area. By focusing on hyperlocal deployments, we can optimize safety and time to deployment by precisely targeting the most relevant aspects for a given area. We iterate on the system by continuously verifying against these requirements and addressing any gaps in performance.

Verification Methods

To verify that our technology meets the defined requirements, we employ various methods that provide efficient and high-confidence validation. These methods include re-simulation of on-road logs, closed course testing, synthetic simulation, and virtual worlds created from existing maps. By using a combination of real-world testing and simulation, we ensure that our system is extensively evaluated and robust in handling diverse scenarios.

Scaling Autonomy for Mass Deployments

As we aim to Scale our deployments to thousands of vehicles across multiple cities, we have designed our technology to facilitate this expansion.

Scaling ML for Rapid Progress

A key factor in scaling autonomy is scaling machine learning. Our ML models continuously evolve as we Collect and mine data from various sources, including on-road logs, targeted data collection, synthetic data, and simulations. This iterative process allows us to identify and address issues efficiently, ensuring ongoing improvement in our system's capabilities.

Building a Holistic Validation Process

Scaling autonomy not only involves scaling the technology but also the entire business infrastructure. We have developed software and operational solutions that enable the remote operation, maintenance, and fleet management of our vehicles. This holistic approach ensures that our fleet operates seamlessly and minimizes human touches, enhancing efficiency and scalability.

Building a Scalable Business

Beyond the technology, we have focused on building a scalable business that provides an exceptional customer experience.

Operating and Maintaining the Fleet Remotely

Our third-generation vehicles are designed to be operated and maintained remotely with minimal human intervention. We can remotely start, shut down, diagnose, troubleshoot, and update the software of our vehicles without physical interaction. This remote management capability enables us to efficiently operate a large fleet and ensure Timely deliveries.

Seamless Integration with Retailer Order Systems

To make the customer experience seamless, we have integrated our platform with retailer order systems. Customers can easily place orders, and our platform ensures smooth coordination and communication between the vehicle and the retailer. This integration streamlines the delivery process and adds convenience for both customers and retailers.

Customer Experience: Ordering, Tracking, and Delivery

Our platform offers customers the ability to place orders through intuitive interfaces on mobile and web platforms. Once an order is placed, customers can track the progress of their delivery in real-time, providing transparency and convenience. Upon arrival, our vehicles are equipped with touch screen interfaces and two-way audio communication, enabling easy and efficient interaction with the customer.

Ensuring Security

Security is of paramount importance in our operations, and we have implemented robust measures to protect our systems and data.

Trusted Vehicle Platform and Teleoperation System

Our custom vehicle platform is designed with security in mind. With secure hardware, cryptography, and a teleoperation system, we ensure that our vehicles are always under our control. Physical protection measures are also implemented to prevent tampering or unauthorized access.

Protecting Partner Data and User Information

We take great care in protecting the data and privacy of our partners and users. We implement technical controls such as encryption and data segregation to safeguard sensitive information. Additionally, we have operation controls in place to prevent compromises and data leaks, ensuring the security of all stakeholders' data.

Emulating Attacks to Verify Security

To ensure the effectiveness of our security measures, we simulate and emulate attacks from potential adversaries. By subjecting our systems to rigorous testing, we can identify vulnerabilities and strengthen our defenses. This proactive approach to security ensures that our autonomous delivery operations remain secure and resilient.

Conclusion

In conclusion, our autonomy stack and custom vehicle are designed to deliver safe, efficient, and scalable autonomous goods delivery. With a rigorous validation process, a focus on scaling both technology and business operations, and a strong emphasis on security, We Are poised to revolutionize the way goods are delivered. Our customer-centric approach and dedication to continuous improvement serve as the foundation for our mission to enhance everyday life through robotics.