Why HD Maps Refuse to Disappear from the Autonomy Stack

The claim that autonomous driving can dispense entirely with high-definition maps has gained currency as end-to-end AI models have grown more capable, yet Here Technologies' head of automotive solutions, Remco Timmer, contends that no credible system is genuinely mapless. Speaking to WardsAuto, Timmer frames the industry's real trajectory as 'map-smart' rather than map-free, arguing that point-to-point autonomy requires a navigable route, awareness of unsignposted rules, and knowledge of speed limits and lane transitions that sensors alone cannot perceive. Even systems marketed as having shed the high-definition layer, he suggests, continue to rely on some form of map data within their sensor fusion.

For automotive leaders, this matters because it reframes a procurement and architecture question that has often been treated as binary. Timmer's argument is that maps provide foresight, context, redundancy and operating boundaries that become essential precisely when perception is compromised, whether by dense urban traffic obstructing sensor fields of view, sun glare degrading camera performance, or European speed regulations that are not consistently signposted beyond a derestriction sign. In each case, the map is positioned not as a crutch for weak perception but as a structural input to sensor fusion, allowing systems to weight camera, radar, lidar and map data differently depending on conditions.

The nuance worth noting is that Timmer does not dismiss end-to-end AI's contribution; he sees it improving scalability, adaptability and more natural driving behaviour, while acknowledging that most engineers are converging on hybrid architectures combining multiple trained models connected through rule sets rather than a single black box. He also points to AI's growing role within mapping itself, with automated updates increasingly supplemented by crowdsourced vehicle sensor data, tempered by human oversight in training, tuning and curation. This suggests the map-versus-mapless framing understates how intertwined the two approaches have become in practice.

What deserves close attention going forward is how the industry resolves what Timmer describes as an unsettled question: the optimal mix of sensors, compute, logic and map data, which currently varies significantly by developer and sensor stack. As regulators press for validation evidence and robustness against edge cases, the case for map-grounded simulation environments and continuous lane modelling may carry weight in safety-case development, particularly for players pursuing point-to-point autonomy at scale. Automotive leaders evaluating AV supplier claims should treat 'mapless' as a marketing distinction rather than a technical one, and probe instead how map data is actually being used within sensor fusion.