Mathematical Models TRIZ Cases

LiDAR point clouds are split into ground and non-ground objects to detect dangerous driving in real time and warn before accidents.

Lane-by-lane traffic jam probability from camera, radar, and lidar data helps vehicles identify less congested lanes in real time.

Actively selected control settings and feedback-updated causal models improve manufacturing yield, cost, and robustness to environmental changes.

Combining signal-light timing with 3D vehicle orientation helps autonomous driving systems infer turning intent despite occlusions and angled views.

Real-time performance feedback updates causal models and control parameters to adapt faster to changing environments with lower resource use.

Machine learning predicts subregion charging demand and clusters hotspots to place mobile EV charging units where queues and recharge delays are highest.

Recorded state transitions update Markov decision probabilities, helping automated vehicles and robots adapt maneuver planning to real environments.

Monolithic laser and lens arrays align optical power into fibers for photonic accelerators while reducing hot spots and non-linear effects.

A learned path model uses recent road segments and historic driving traces to predict junction choices more accurately, even on non-routine trips.

Scoped sensor histories exclude failure-period noise so classification models can detect precursor states and alert on likely component failures.

Histogram and probability-model analysis isolates critical vehicle scenarios from limited measurement data to support reliable SOTIF validation.

Recent radar updates are combined with a full perception pipeline to keep object tracks current and cut autonomous vehicle latency.

Monte Carlo tree search predicts uncertain future driving states to improve decision robustness beyond DQN while limiting computation.

Machine learning reweights biased autonomous vehicle simulation scenarios to match ODD exposure and estimate real-world performance metrics accurately.

Environmental sensors raise or lower vehicle event detection thresholds to capture anomalous events more reliably under light, noise, and intersection distractions.

Vehicle sensor data drives reinforcement learning and neural network alignment control to adapt camber and toe for fuel economy, tire wear, and stability.

A causal control model updates internal parameters from live environment responses to adapt quickly, cut computation, and keep settings safe.

HD map localization auto-generates ground truth labels from sensor data, cutting manual annotation time while improving DNN accuracy.

Surprise-based evidence accumulation predicts when traffic agents react by modeling unexpected trajectory changes and stimulus onset.

Multiple chips infer sensor index values and an anchor chip aggregates them for real-time vehicle driving control with manageable complexity.

Cluster-specific causal models speed control learning in changing environments while keeping exploration efficient and within safe ranges.

A contour-constrained compound radar model adapts online to track both kinematic and extended object states more accurately in autonomous driving.

By removing non-safety-critical states, D2RL creates realistic AV test traffic that speeds safety validation in complex driving environments.

Fault rates from key vehicle parts enable accurate failure detection and selective autonomous function shutdown with less data processing.

A trained prediction model infers valve element position from coil signals, avoiding hardware sensors while handling hysteresis and temperature effects.

Predicts lane-based trajectories for road users who cannot see each other, then triggers alerts only when collision probability is high.

Measurement feedback and AI update wafer etching recipes in real time to keep process parameters in range and protect yield and reliability.

Injection-locked VCSEL slave lasers use optical feedback and polarization rotation to solve Ising optimization with high speed and low power.

Causal learning links vehicle control signals to measured parameters, enabling faster adaptation with lower computational load and less wear.

Deep learning infers air resistance and friction from vehicle data, enabling faster, more accurate real-time driving control.

Localizing sensor data to HD maps automates ground truth labeling, cutting manual effort while improving autonomous DNN training accuracy.

Onboard sensors and image processing detect adverse vehicle events, generate FNOL claim data, and route it automatically with less human intervention.

Neural network clustering detects anomalous vehicle trips more accurately than fixed thresholds while reducing computing and network load.

Learned perturbation models simulate correlated lane detection errors from ground truth, cutting AV safety testing cost and simulation complexity.

Search tree pruning and reward propagation cut fleet control computation while helping shared-resource vehicles avoid deadlocks.

A state-wide Bayesian filter iteratively calibrates complex digital twin parameters from one common observable, reducing ambiguity.

Transformer-based scene risk prediction handles multimodal sensor data directly, cutting pipeline complexity for real-time autonomous vehicle control.

RNN and CNN models estimate Li-ion battery SOC and SOH from voltage, current, and temperature time series with lower real-time overhead.

Sensor fusion and predictive modeling turn lane traffic density, flow, and speed into real-time jam probability for smoother assisted driving.