October | 2020
Introduction
An autonomous shuttle is a vehicle that moves autonomously at speeds lower than 50 miles per hour on pre-charted routes under remote surveillance and environment restrictions for operations1 . All autonomous vehicles which fall under this category are electric, used to ferry people or deliver goods, and manned or unmanned. In any case, they always have a remote operator, who is one of the safety fallback mechanism enablers. Autonomous shuttles are now moving from being remote operated to remote assisted, depending on the Operational Design Domain (ODD) and the extent to which human intervention is anticipated. Given the constrained environments under which an autonomous shuttle is deployed, they act as an enabler to fast-track autonomous technology development before largescale deployment.
This article takes the reader through the technology enablers for navigation of autonomous shuttles, highlighting, in the first half, how autonomous shuttles will accelerate self-driving technology perfection steering through regulatory grey areas. The second half of the paper talks about the future of these vehicles, and areas of potential fine-tuning.
Figure 1. Possible deployments of Autonomous shuttles
Aspects of Autonomous shuttles
Shuttles as described by Regulatory authorities
LEV (Low Emission vehicles), unmanned vehicles (in some cases), LSV (low speed vehicles) are different technical terminologies associated with shuttles. The NHTSA is rather stringent about the weight and dimensions of Low Speed Vehicles. Per the NHTSA, a Low Speed Vehicle is “a motor vehicle, (1) that is 4-wheeled; (2) whose speed attainable in 1.6 km [kilometers] (1 mile) is more than 32 kilometers per hour (20 miles per hour) and not more than 40 kilometers per hour (25 miles per hour) on a paved level surface, and (3) whose GVWR [gross vehicle weight rating] is less than 1,361 kilograms (3,000 pounds).”2
Key enablers for Autonomous shuttles
Depicted in figure 2 are the basic facets of autonomous shuttles. They require pre-charted maps and well-defined routes. They also require geography-specific customizations like identifying common objects and understanding the local traffic laws and regulations. The complexity of the driving scenario to which an autonomous shuttle is exposed is limited, and they have well-defined emergency protocols. The design domain in which an autonomous shuttle operates is highly restricted, for instance, by design, they are not meant to share road space with faster moving traffic.
Figure 2: Aspects of Autonomous shuttles and concerns
The poised technology and regulatory stance
Autonomous shuttles are the sweet spot for self-driving technology deployment with minimal chances of on-road fatalities and continuous technology improvement. The vehicle size is perfect, making it fall under the LSV category, with potential for regulatory relaxation as detailed in the case below.
Nuro R2X is Nuro’s second-generation unmanned and occupant-less delivery vehicle. In February 2020, Nuro managed to get certain exemptions from the NHTSA & FMVSS for the R2X vehicle, on the premise that the vehicle is occupant-less3. Firstly, removal of exterior and interior rear-view mirrors; secondly, doing away with real-time display requirements for the rear-view camera; lastly, on the usage of glazed material windshield that was meant to protect occupants. This, in turn, makes R2X safer for pedestrians. R2X also boasts a crumple zone in the front-end design, thereby reducing the impact on a road user.4
Safety standards and testing protocols for automated functions aren’t matured in the shuttles space. Since autonomous shuttles do not have pedals, steering wheels, they do not comply with NHTSA guidelines.5
Mapping for Autonomous shuttles
Autonomous shuttles move on pre-mapped routes. The route that these vehicles are supposed to follow are stored locally in the shuttle and accessed to make the vehicle move predictably, stopping only at boarding or de-boarding stations or for emergency stops.
The operational specifics of the shuttle are also decided based on this route. Any autonomous shuttle needs to precisely locate itself on this pre-defined map. This is known as localization and is achieved by prior knowledge of landmarks and real-time sensor observations made by the vehicle. As part of the deployment checklist, if the experts from the autonomous shuttle company so deem, new features are “installed” on the route chosen to ease localization.
The operations team has to ensure safety of the occupants as well as road users who are a part of the dynamic environment. Events like diversions owing to accidents or temporary construction zones etc. will have to be controlled by the remote operations team. Any detour requires surrounding road network knowledge and that may not always be the case. Per Optimus Ride’s CEO, a “geo-fenced strategy” is being used for Autonomous shuttles.6
Detailed mapping data of the intended route is saved locally in onboard storage. This helps with unanticipated events that trigger the safe-stop mode of the shuttle.7
Restricted Operational Design Domain
Deployments of Autonomous shuttles so far have happened in constrained and more protected environments. As depicted in figure 1, these environments include university campuses, hospital premises, closed corporate parks, technology parks, and some pre-determined routes for carrying people with known boarding and de-boarding areas. Within these enclosed spaces, too, is a spectrum of complexity. An airport deployment use-case for movement of people from security to boarding gates is a less complex ODD as compared to in-city driving for delivery of goods, which is almost on the other extreme of the spectrum.
Emergency handling
Every critical system in autonomous shuttles like acceleration, braking, sensors, computing, and steering systems have built-in layers of redundancy; in most cases, they have triple redundancy. Autonomous shuttles are at level 3/4 of autonomy per SAE (depending on the application). Level 4 autonomous vehicles are constituted by ISO26262 described ASIL C or ASIL D components at the hardware & software level.
The vehicle is designed to come to a safe stop when it exits its intended ODD. For example, if an autonomous shuttle is designed for wet roads, but not for icy or snow-covered roads, then the remote operator can intervene if such a scenario arises.
The onboard computer in these shuttles is programmed to compute the trajectory to be taken for a safe stop, keeping in mind various dynamic environment factors like traffic/obstacles in the immediate vicinity.
Along with Product Safety, Deployment safety and Operational safety are two major focus areas where autonomous shuttle companies ensure safety standards are adhered to.8 As part of deployment safety, an onsite team inspects the route and gets it ready for shuttle deployment. The operations team may also have an onsite pre-trip checklist, an Operations Manager along with a team who is contacted in case the shuttle comes to a safe stop.
Staying relevant in the COVID era
COVID-19 has had people stay put where they are. There has been a shift from people movement to goods movement. This shift is seen on the employment side, with people employed in taxi or cab services moving to goods delivery. The shift also means implications for autonomous shuttle companies, and they are repurposing shuttles meant for movement of people into goods carriers. There is a last-mile delivery boom, with Uber Eats search increasing by 70%, DoorDash 55%, Grubhub 46%, and Postmates 42%9. With this boom, Optimus Ride, Beep, Gatik, and Cruise are now delivering food10.
Future readiness:
This section talks about the infrastructure capabilities required for Autonomous shuttles and some aspects that require improvement.
Dedicated lanes:
Are autonomous shuttles to stay with us for long? The accelerated deployment of autonomous shuttles is winning the autonomous machines’ ecosystem the necessary opportunity for eventual technology perfection, so autonomous shuttles apparently are the intermediate step in autonomous technology real-world deployments. If this is the case, are we creating “room” for them to “thrive” alongside passenger vehicles?
For reasons similar to why trucks are a threat to passenger vehicles on highways, passenger vehicles will be threats to these shuttles. The weight of trucks could be as much as 20-30 times the weight of other passenger vehicles.11 The weight of passenger vehicles is typically between 2,000 to 5,000 pounds, but could be greater for light trucks and vans (LTVs)12. The weight of passenger vehicles is hence more than twice that of LSVs. This puts autonomous shuttles at greater risk during collision and consequent fatalities with any other motor vehicle on the road.13
A good approach? Creating extra room for autonomous shuttles, at least wherever there is scope. Wherever smart cities or smart campuses are being made, dedicated lanes could give them multiple advantages (refer figure 3). Currently, one other constraint in their deployment is that they can share driving space only with slow-moving traffic. With a dedicated lane, the possibility is that since all autonomous shuttles moving at a low speed will use a dedicated lane, they will be safer.
Figure 3. Today roads have dedicated bike lanes, in the future dedicated autonomous shuttle/LSV lane (here next to bike lane) could imply more deployment (Left-handed driving system)
Shuttle speed and reaction times
All Autonomous shuttles are intended to operate at low speeds. The operation speed chosen gives the vehicle an added advantage of more time in taking certain decisions. A crude back-of-the-envelope calculation suggests that – if a vehicle was operating at 72 kmph (20 metres per second), then a camera with a frame rate of 40 fps would imply that in the time taken for one frame to be captured (1/40th of a second), the vehicle will have moved by 50cm. Add to it the processing time, and the vehicle may move by a few more centimetres before a decision is taken.
Explained above cannot be a good enough reaction time, meaning sensors and processors need to have better frame rate, processing powers, or the vehicle needs to be slower. However, one needs to know that the slow speed prepares the vehicle, for any events which are “recorded” or “captured” by the vehicle, well in advance. There is also a fair chance that an event with an element of surprise, for instance, a child chasing his/her pet dog on the road right in front of the shuttle, will not be able to get the right response from it. Essentially, there is a region beyond which the shuttle with its low speed, will be able to respond to all kinds of events, sudden or stochastic.
Object avoidance, object classification:
All autonomous machines are expected to have perfect object-detection capabilities. However, object-classification capabilities are subject to familiarity with geography-specific object classes. This in turn will need re-training of object-classification algorithms using data acquired in the local region. Object-classification capabilities of autonomous systems are hence near perfect.
When it comes to the reaction of an autonomous machine toward objects, most systems are designed to freeze in their tracks when an object blocks them. A few autonomous systems have the intelligence to predict the behavior of these objects and chart an alternate object avoidance trajectory for themselves. The object-avoidance capabilities of autonomous systems are also subject to application and ODD. For instance, in a completely robotic environment within a warehouse, autonomous systems can do without these cognitive capabilities whereas a co-bot requirement will need this decision-making capability.
Object avoidance has to be a necessary feature in autonomous shuttles, which is not the case today.
Autonomous shuttles - Business dynamics:
Most Tier 1 companies have close-knit buyer-vendor relationships with Automotive OEMs. However, Autonomous shuttles is a space that is attracting investments, partnerships, and acquisitions from Tier 1s. Autonomous shuttles have been flourishing in partnerships with these Tier 1 automotive companies. For instance, EasyMile has partnered with Continental14, similarly ZF has acquired 2getthere15, Nuro also has a Tier 1 automotive technology partner whose name hasn’t been disclosed. With the advent of autonomous technology, and the leadership in this space by disrupters like Waymo and Baidu, Tier 1 companies are becoming redundant. Working in partnership with new OEM categories like autonomous shuttle companies is helping them vertically integrate.
Conclusion:
Autonomous shuttles is a world of stepwise autonomous technology introduction and improvement in chunks. Per market reports, the Global Low Speed Autonomous Driving Market is expected to grow from USD 2.5 Billion in 2019 to USD 7.3 Billion in 2025 at a CAGR of 19.57%16. While other Autonomous vehicles struggle with targets of zero accidents and reduction of disengagement scenarios, autonomous shuttles are getting deployed with operational monitoring support to match safety standards. They are, undoubtedly, in the nascent stages of evolution, but here to stay.
References:
Balaji Sunil Kumar
Sunil currently leads the AI Algo Stack team for Wipro’s Autonomous Systems and Robotics practice. He is involved in developing Algorithms stack related to Automated Guided Vehicle systems. Sunil is also a Senior Member of Wipro’s DMTS community and has around 22 years of experience working primarily with Embedded Systems across a variety of industries. He can be reached at balaji.kumar@wipro.com.
Sreesankar R
Sreesankar is a Senior Architect in the Wipro Autonomous Systems and Robotics practice team. He has over 19 years of experience in varied domains like Automotive, Embedded, and Securities. Currently, he leads the Wipro Auto Annotation Studio team, focusing on Computer Vision, Sensor fusion, Machine Learning, and Explainable AI, with the objective of managing and improving the AI system lifecycle. Sreesankar can be reached at sree.sankar@wipro.com.
Garima Jain
Garima is a Global Business Manager in Wipro’s Global 100 leadership program, working across functions and business units in India and the US, with rotations in pre-sales, sales, delivery, and domain consulting. Prior to completing her MBA from the Indian Institute of Management in Bangalore, she worked as an engineer on automotive core chip design for three years. She is passionate about robotics and all things related to autonomous and connected vehicles. Garima can be reached at garima.jain5@wipro.com.
This white paper explores the drawbacks in existing QFR technique and also proposes solution which can improve QFR success rate.
How a comprehensive framework for efficient network testing can help operators deploy LTE networks faster
This paper explores the possibility of eMBMS becoming the standard technology platform for transmission of multimedia content over cellular networks. It details out the advantages of adopting eMBMS and the various scenarios in which it can play a game-changing role
Locations
