Leveling Up in Lockstep with the Evolution of Autonomous Driving
Note: Woven Planet became Woven by Toyota on April 1, 2023. Note: This blog was originally published on October 22, 2020 by CARMERA, which is now part of Woven Planet. This post was updated on December 13, 2022.
By Ro Gupta, VP of Automated Mapping
I was asked to present a keynote last week at the McKinsey Center for Future Mobility’s autonomous driving roundtable to talk about the evolving role of maps. With experts drawn from across the auto-tech landscape, the event presented a unique opportunity to discuss what problems maps have been focused on to date, where that focus is shifting now and what additional applications we’ll see in the future.
Earlier this year, we found it useful to borrow from Maslow’s famous hierarchy of needs to develop a sequencing framework that reflects the various considerations we’ve heard from partners who think holistically about commercializing autonomous driving at scale. We decided to share a version with the roundtable, and now here in this post, both to describe the increasing requirements and utility for/from maps over time, and to offer a generalizable way to think about other problems that call for hierarchical systems thinking.
The role of maps as a critical safety layer in the AV tech stack is now fairly well understood. HD maps with highly detailed, accurate representations of road features and geometries have become gating factors to help AVs ensure they understand their spatial position and what’s around them, earning them the moniker the “4th sensor.”
But in talking to key stakeholders — including OEMs, integrators, chip makers, dealers, insurers and regulators — we found that thinking of maps purely within the safety context provided too narrow a view of the role maps are expected to play going forward.
Safety is the foundation and always will be, but on top sits a pyramid of advanced needs. These can be grouped into five core categories, each category building on the prior — though in some cases we’re likely to see parallel development as well:
Safety — the need to meet the core redundancy and foresight requirements;
Experience — the need to provide a consistent, enjoyable ride;
Efficiency — the need to task compute and connect resources efficiently;
Economics — the need to support new models of monetization;
Compliance — the need to create transparency with underwriters and policy makers.
The “Mapping Hierarchy of Needs”: Safety → Experience → Efficiency → Economics → Compliance
Each layer of this hierarchy calls upon a different dimension of the utility of map data, which, in turn, may have differing requirements in terms of detail, accuracy or freshness.
To illustrate, let’s look at Experience, focusing in this case on privately owned vehicles. We think of user experience as having three sub-dimensions:
Availability — will the autonomous driving feature be consistently and predictably available?
Smoothness — will the autonomous driving feature provide a smooth, on-rails driving experience?
Optimality — will the autonomous driving feature reach the desired destination as quickly and efficiently as possible?
The UX Continuum: maps are expected to contribute to increasing levels of expected user experience.
These dimensions build on each other, like a UX ladder. Today, most customers are on the first rung, simply wanting a feature that’s consistently available. The biggest critique of GM’s otherwise highly lauded Super Cruise feature, for example, is its inconsistent and/or unpredictable availability, largely a function of whether a given road has been mapped and is not currently under construction.
But as vehicles become more sophisticated and choices become more plentiful, customers will become ever more demanding, wanting a better experience in more places. A frequently cited and useful analogue is the airline industry. There was a time — I can even remember myself as a kid — when basic safety was top of mind when considering flying. Today, however, customers decide almost exclusively based on price, on-time expectation, convenience and quality of service.
In terms of data requirements, each rung of this UX ladder implicates an increasingly advanced layer of map capability with more demanding refresh rates.
Climbing the UX ladder, the change in map function necessitates a change in the type of primary map data and the rate at which that data is refreshed.
Availability is essentially a question of what’s required for robust localization. All things equal, an automated driving feature where the human is still expected to take back control will be available if the map has enough information to allow the vehicle to accurately determine its position. HD maps are relatively resilient in this regard — a single missing signpost usually won’t throw the vehicle off, so a map with a less frequent update cycle can suffice.
Smoothness, on the other hand, relies more heavily on navigational vectors — the “virtual rails” that an autonomous vehicle expects to be able to follow. A discrepancy between these vectors and lane-level reality — for example, an unexpected construction zone — can cause abrupt experiences such as slowing, pausing and/or disengaging as the vehicle attempts to negotiate unexpected changes to its drive path in real-time. This layer of map data, therefore, is far more susceptible to change, and requires a much more frequent update rate.
To achieve an Optimal ride, particularly for unsupervised situations when humans are not expected to be in the loop, vehicles will require more sophisticated path planning capability to minimize ETAs instead of just routing around risky areas. This includes pulling in more ephemeral data such as temporary closures, emergency vehicle activity, micro-weather events and so on, which approach the upper bound of update cadence, in some cases, near real-time.
Establishing this “hierarchy within a hierarchy” provides a basis for a phased rollout anchored to meeting each UX standard over time. Factors influencing this phasing include which roads have the highest average vehicle-miles travelled (VMT) and serve the largest shares of population and/or customer base for a particular brand. Regardless of the specific inputs, this common set of variables helps stakeholders set expectations and the industry level-set on desired capabilities and feasibility over time:
Current state: OEMs and suppliers are getting specific on aligning data refresh standards, cost implications and customer experience expectations.
The UX story is just one slice of the Hierarchy of Needs, but it illustrates how we think about the expanding, evolving role of maps and the ever-growing importance of change management in the mapping stack.
We’re looking for talented engineers to help us on this journey. If you’d like to join us, we’re hiring!