Autonomous Driving Patent Landscape for Occupancy Network Architectures

Overview of Technical Issues:

The request seeks patent landscape intelligence for occupancy network architectures in autonomous driving, where insufficient visibility into existing IP protection creates uncertainty about freedom-to-operate and design constraints; the goal is to map the patent terrain to identify protected technical approaches, key patent holders, and available design spaces for safe commercialization.

Solution directions generated for this problem

Problem Direction 1 :

ImprovePatent search coverage completeness
VS
ConstraintPatent analysis resource consumption

Inspiration 1 : Cross-domain reference

Application Principle: #35 Parameter changes
Cross-domain applicability Assess applicability
Surgical System Leveraging Large Language Models
Innovative Solution Refine solution

Risk-stratified adaptive search intensity allocation for patent landscape mapping

Stratify patent landscape by litigation risk density
How to solve :
  • Divide occupancy network domain into risk zones using citation density (≥50 forward citations = high-risk) and litigation history (active cases in past 3 years)
  • allocate 70% of expanded search budget to top 20% high-risk zones (sensor fusion, 3D convolution), 20% to medium zones, 10% to low-risk areas
  • Deploy automated risk scoring using ML classifier trained on 500+ annotated patents—flags high-risk families with 85%+ precision in initial 48-hour scan, triggering deep multi-jurisdiction search (USPTO, EPO, CNIPA) only for flagged clusters
  • For low-risk zones, maintain baseline 60% coverage using single-database keyword search, accepting controlled blind spots in non-core areas
  • Quality gate: weekly risk map review—if new high-citation patent emerges in "low-risk" zone, immediately re-classify and allocate resources
Expected Effect : 95% coverage in critical zones with 2.2× resource vs 4-5× uniform expansion; litigation risk capture rate ≥92%
Risk Control :
  • risk classification model drift over time
  • emerging technical sub-domains misclassified as low-risk
  • cross-jurisdictional citation data incompleteness

Problem Direction 2 :

ImproveClaim boundary interpretation accuracy
VS
ConstraintFreedom-to-operate assessment duration

Inspiration 1 : Cross-domain reference

Application Principle: #1 Segmentation
Cross-domain applicability Assess applicability
Far-field extension for digital assistant services
Innovative Solution Refine solution

Modular parallel claim analysis with independent technical cluster segmentation

Divide occupancy network patents into independent technical modules for parallel processing
How to solve :
  • Segment patent portfolio into 5-7 independent technical clusters (voxel encoding, transformer architectures, sensor fusion, temporal aggregation, occlusion handling, multi-view geometry, output decoding) using automated claim element extraction within 2 days
  • Assign dedicated 2-analyst teams to each cluster for parallel deep claim construction analysis — each team processes 15-20 patents with <5% ambiguity target using standardized claim chart templates and file history review protocols over 3 weeks
  • Implement cross-cluster conflict detection in final week using feature overlap matrix to identify boundary patents spanning multiple clusters, with senior analyst reconciliation sessions (4 hours per conflict case)
Expected Effect : Assessment duration compressed to 4-5 weeks; interpretation accuracy <5% ambiguity maintained; analyst utilization efficiency +60%
Risk Control :
  • cluster boundary definition ambiguity causing patent misallocation
  • inter-cluster dependency patents requiring duplicate analysis
  • team capability variance affecting accuracy consistency

Problem Direction 3 :

ImproveIP landscape mapping granularity
VS
ConstraintFreedom-to-operate assessment duration

Inspiration 1 : Cross-domain reference

Application Principle: #17 Another dimension
Cross-domain applicability Assess applicability
Electronic cigarette having dual air passages
Innovative Solution Refine solution

Spatial-layered patent feature mapping with progressive depth allocation

Spatial layering transforms patent landscape into multi-dimensional feature space
How to solve :
  • Map occupancy network patents into 3D spatial coordinates (X-axis: input modality [camera/LiDAR/radar], Y-axis: spatial representation [voxel/BEV/point cloud], Z-axis: temporal aggregation [single-frame/recurrent/transformer]) using automated claim parsing within 5 days, generating visual clusters without sequential patent-by-patent analysis
  • Depth-stratified analysis allocates resources by spatial density—high-density zones (≥8 patents per cubic unit) receive full claim construction (12 hours/patent, <5% ambiguity), medium-density zones (3-7 patents) receive element-level mapping (4 hours/patent, ~12% ambiguity), sparse zones (<3 patents) receive automated tagging only (0.5 hours/patent)
  • White space corridors auto-identified as unoccupied coordinate regions with ≥2-unit clearance from protected clusters, validated through cross-sectional claim element checks on boundary patents, delivering actionable design freedom map in 4-5 weeks total
Expected Effect : Granular feature mapping in 4-5 weeks vs 8-12 weeks baseline; white space identification accuracy ≥90%; resource efficiency +60%
Risk Control :
  • coordinate axis definition subjectivity
  • automated claim parsing accuracy <85% for complex dependent claims
  • spatial clustering misses functional equivalents across coordinate boundaries

Problem Direction 4 :

ImproveClaim boundary interpretation accuracy
VS
ConstraintPatent analysis resource consumption

Inspiration 1 : Cross-domain reference

Application Principle: #27 Cheap short-living objects
Cross-domain applicability Assess applicability
Aggregation of unique user invocations in an online environment
Innovative Solution Refine solution

Tiered probabilistic claim filtering for resource-efficient patent analysis

Deploy probabilistic claim filters for rapid triage
How to solve :
  • Implement Bloom filter-based claim element matching — hash each occupancy network design feature (voxel encoding, sensor fusion architecture, temporal aggregation) into 3-5 hash functions, compare against patent claim elements to flag potential overlaps in <2 hours per patent with 15% false positive rate acceptable for initial screening
  • Route negative-match patents (70% of corpus) directly to low-risk category with automated 1-page summary, consuming 0.5 analyst hours each
  • Escalate positive-match patents (30% flagged) to full claim construction analysis with file history review, investing 8 hours per patent to achieve <5% ambiguity only where infringement risk exists
Expected Effect : Average resource per patent reduced from 8 to 3.1 hours; 95% coverage achieved within 2.5× baseline effort instead of 4-5×
Risk Control :
  • Bloom filter false negative rate causing missed infringement risks
  • hash function design inadequate for complex claim language
  • analyst skill variance in escalation threshold calibration
Patsnap Eureka Solution