BEV_FUSION_V3
Research · Safety · Vehicle Intelligence
Researching the systems behind safe autonomous mobility.
We develop the mathematical, perceptual, and control foundations required to transition autonomous driving from isolated neural network models to provably safe, deployable vehicle software.
Autonomous Perception
Robust environmental representation across surround cameras, solid-state LiDAR, and millimeter-wave radar under heavy adverse weather and dense urban occlusion.
VOXEL_OCC_3D
Adverse Weather Occupancy Grids via Multi-Modal Temporal Voxels
LANE_TOPOLOGY
Semantic Road Geometry Extraction Without Pre-Mapped HD Anchors
Prediction & Agent Forecasting
Modeling uncertainty over extended time horizons, multi-agent game-theoretic road dynamics, and pedestrian intent under unscripted traffic conditions.
PROBABILISTIC_TRAJ
Multi-Modal Agent Trajectory Forecasting via Latent Graph Diffusions
INTENT_GAZE_DMS
Occluded Pedestrian Crossing Intent Recognition with Head Pose Priors
NASH_MERGE_SOLVER
Equilibrium Solutions for Highway Bottleneck Ingestion & Merge Negotiation
EXPANDED_TURNS
Expanded Moving-Turning Stratum: Simple Rule vs Neural Network
EVIDENCE_ABSENCE
Evidence of Absence: Heading Change & Neighbor Cues in the Last 2 Seconds
HEURISTIC_TREE
History-Only Tree vs Simple Rule: Milder Arcs on Already-Bending Turns
Planning & Predictive Vehicle Control
Synthesizing collision-free, dynamically feasible trajectories with formal safety guarantees and real-time sub-millisecond MPC solvers.
CBF_SAFETY_BARRIER
Control Barrier Functions for Deterministic Collision Avoidance in Unstructured Roads
MPC_400HZ_RUNTIME
Sub-Millisecond Non-Linear Model Predictive Control for Highway Evasive Steering
ASIL_D_MRM
Minimum Risk Maneuver (MRM) Synthesis Under Sensor Degradation Cascades
Domain 04
Localization & SLAM
Sub-decimeter semantic localization operating reliably across urban tunnels, multi-level parking garages, and GPS-deprived canyons.
Domain 05
Vehicle Intelligence & OS
Deterministic microkernel architectures, AUTOSAR Adaptive bindings, zero-copy IPC, and OTA resilience pipelines for mass-market EVs.
Domain 06
Safety & Formal Validation
Scenario-based digital twin validation, ISO 26262 ASIL-D compliance cases, and adversarial simulation stress frameworks.
Layer 3 Authority · Systems & Verification
Research that moves from benchmark models to real-world vehicle software.
We validate every algorithmic release against hardware-in-the-loop (HIL) dynamometer rigs, CAN/Ethernet timing jitter limits, and rigorous safety barrier conditions before code reaches fleet vehicles.
< 14.2ms
End-to-end perception to actuation latency
99.999%
ASIL-D fail-safe watchdog guarantee
4.8M km
Closed-loop adversarial sim coverage
Featured Research Roles
Perception Lab
Research Scientist — Multimodal BEV Fusion
Design foundation models that fuse asynchronous radar, camera, and ultrasonic feeds in real-time.
Planning & Control
Robotics Engineer — Non-Linear MPC
Formulate real-time convex optimization and Control Barrier Functions for autonomous collision evasions.
Systems Architecture
Staff Engineer — ASIL-D Embedded RTOS
Architect deterministic microkernel schedulers and hardware redundancy systems for EV drive-by-wire.
Explore all Svashasan research
Explore our publications, patents, technical reports, safety cases, and dataset milestones across the entire autonomous vehicle stack.
Research Index
A chronological library of papers, technical reports, system releases, and safety verifications.