Multi-sensor fusion platform for bootstrapping the training of a beam steering radar

This patent covers a multi-sensor fusion architecture for autonomous vehicles in which three parallel perception engines — a camera perception engine built on a camera neural network, a lidar perception engine built on a lidar neural network, and a radar perception engine built on a radar neural network coupled to a beam steering radar — each detect and identify objects in their respective sensor data streams. The novel core of claim 1 is the training relationship: a system controller is adapted to train the radar neural network using the training of the camera and lidar perception engines. In plain terms, the camera and lidar networks — modalities that are easier to label and train against — are used to "bootstrap" the harder-to-train radar network. This addresses a practical pain point in automotive radar perception: ground-truth labeling of radar returns is difficult, so the patent leverages the better-understood optical and lidar pipelines as a teaching signal for the radar pipeline.

The patent is assigned to Metawave Corporation, with Matthew Paul Harrison as the sole named inventor. It was granted in 2023 under primary CPC Radar & Navigation (G01S). The family is small — 2 members in a single country — indicating a focused, domestic filing strategy rather than aggressive international prosecution. Forward citations stand at 3, which is low; for a grant this recent (2023), that is expected and likely reflects publication lag rather than lack of significance — citation-weighted ranking systematically under-rewards fresh grants.

Where this patent stands: This is a niche, technically specific filing centered on Metawave's particular commercial focus — beam steering (metamaterial/analog) radar with on-board neural perception. Rather than a foundational sensor-fusion anchor, it is best read as a defensive/product-aligned patent protecting a specific training methodology (cross-modal bootstrapping for radar) tied to Metawave's beam steering radar hardware. Metawave does not appear in the top-assignee list of this space, so the patent sits adjacent to, rather than at the center of, the dominant autonomous-driving perception portfolios held by the larger players. Its value is tightly coupled to the commercial viability of beam steering radar as a sensor class and the specific bootstrapping training approach it claims.

The surrounding space is crowded and tightly clustered around autonomous-vehicle perception, with several CPC areas concentrated far above corpus baseline — most strikingly Autonomous Control (G05D) at 53× corpus baseline, Radar & Navigation (G01S) at 38.6×, Traffic Control (G08G) at 32.8×, and Computer Vision (G06V) at 27.2×. These multipliers confirm a densely populated, specialized field rather than a diffuse one. By citation impact, the dominant players in this result set are Volkswagen AG (971 citations), Mobileye Technologies, NVIDIA (3 patents, 222 citations), and Uber Technologies (2 patents). However, the patents nearest to the subject by relevance are clustered around camera-radar-lidar fusion and neural-network training — a slightly different sub-cluster where Tesla (3 of the nearest 20: US-11215999, US-11150664, US-10956755), NVIDIA, and UATC/Uber appear repeatedly, though none of those nearest-neighbor specialists individually dominate the citation chart the way Volkswagen does on a single high-impact filing.

The closest related patents:

• US-10776673-B2 — Learning method and device for sensor fusion to integrate radar and camera information to improve a neural network for autonomous driving (StradVision Inc · 31 citations · 10-member family). The most on-point neighbor: it directly addresses training a fusion neural network from radar plus camera data, which overlaps heavily with the subject's cross-modal training premise, though it does not specify beam steering radar or three-way camera/lidar/radar bootstrapping.
• US-11688181-B2 — Sensor fusion for autonomous machine applications using machine learning (NVIDIA Corporation · 7 citations · 7-member family). A broad ML-based sensor fusion filing from a citation-leading assignee; relates to the same fusion-plus-learning architecture but at a more general level.
• US-11494937-B2 — Multi-task multi-sensor fusion for three-dimensional object detection (UATC LLC · 4 citations · 5-member family). Covers fusing multiple sensor modalities for 3D detection, overlapping the subject's multi-engine object-detection structure.
• US-11287523-B2 — Method and apparatus for enhanced camera and radar sensor fusion (CMMB Vision USA Inc · 0 citations · 3-member family). A directly on-point camera-radar fusion filing with zero forward citations — strategic IP the citation-weighted score under-rewards.

The top assignees here are predominantly commercial (NVIDIA, Tesla, Uber/UATC, Volkswagen, Mobileye, Texas Instruments, Raytheon), with only a couple of academic entrants (Tsinghua University, plus institute-style filers), so freedom-to-operate against corporate portfolios is the central concern rather than university licensing. No single CPC area is the sole driver, but the combination of Autonomous Control, Radar, and Computer Vision concentrations confirms the subject patent sits inside a tightly clustered, actively prosecuted perception-and-fusion field.