We build a proof of concept for your problem for 5,000–8,000€. You get a tangible demo instead of slides with promises.
Developing a self-driving taxi service in dense urban environments posed immense challenges for Cruise. Complex scenarios like unpredictable pedestrians, erratic cyclists, construction zones, and adverse weather demanded near-perfect perception and decision-making in real-time.[1] Safety was paramount, as any failure could result in accidents, regulatory scrutiny, or public backlash. Early testing revealed gaps in handling edge cases, such as emergency vehicles or occluded objects, requiring robust AI to exceed human driver performance.
A pivotal safety incident in October 2023 amplified these issues: a Cruise vehicle struck a pedestrian pushed into its path by a hit-and-run driver, then dragged her while fleeing the scene, leading to suspension of operations nationwide. This exposed vulnerabilities in post-collision behavior, sensor fusion under chaos, and regulatory compliance. Scaling to commercial robotaxi fleets while achieving zero at-fault incidents proved elusive amid $10B+ investments from GM.[2][3]
Cruise addressed these with an integrated AI stack leveraging computer vision for perception and reinforcement learning for planning. Lidar, radar, and 30+ cameras fed into CNNs and transformers for object detection, semantic segmentation, and scene prediction, processing 360° views at high fidelity even in low light or rain.[6]
Reinforcement learning optimized trajectory planning and behavioral decisions, trained on millions of simulated miles to handle rare events. End-to-end neural networks refined motion forecasting, while simulation frameworks accelerated iteration without real-world risk. Post-incident, Cruise enhanced safety protocols, resuming supervised testing in 2024 with improved disengagement rates.[3] GM's pivot integrated this tech into Super Cruise evolution for personal vehicles.
We build a proof of concept for your problem for 5,000–8,000€. You get a tangible demo instead of slides with promises.
Cruise was founded in 2013, acquired by GM in 2016 for $1B, marking GM's aggressive entry into AVs.[1] By 2018, fourth-gen vehicles without steering wheels debuted, tested in SF and Phoenix. 2022 saw expansion to Austin and Houston with unsupervised rides. Commercial robotaxi launches occurred in SF by 2022, but 2023 incidents halted progress: California DMV suspended permits after the pedestrian drag, NHTSA investigated.[2][3] CEO Kyle Vogt resigned Nov 2023. May 2024 resumed mapped testing; Aug 2024 driverless testing restarted in Phoenix. Dec 10, 2024, GM announced end to robotaxi funding, absorbing Cruise into personal AV development targeting 2028 eyes-off systems.[4][5]
Cruise's perception system fused data from lidar (200m range), radar, and cameras using deep neural networks. CNNs like ResNet variants handled object detection (vehicles, peds at 99%+ accuracy), while transformers enabled bird's-eye-view (BEV) mapping for 3D occupancy.[6] Semantic segmentation via U-Net segmented lanes/curbs. Challenges like night/rain occlusion were mitigated by multi-modal fusion and temporal consistency from RNNs/LSTMs, achieving <1% false positives in benchmarks.
Planning and control relied on reinforcement learning (RL) agents trained in high-fidelity simulations mimicking urban chaos. RL policies optimized reward functions for safety, efficiency, and comfort, simulating billions of miles. Imitation learning bootstrapped from expert data, transitioning to RL for exploration. Motion forecasting used graph neural networks predicting trajectories over 8s horizons. Post-2023, rule-based safeguards overlaid RL to prevent invalid actions, reducing intervention rates to <1 per 10k miles in tests.[3]
5M+ real driverless miles fed datasets for training, augmented by CARLA/NuPlan simulations. Fleet data looped back via shadow mode, validating predictions offline. Compute scaled on GPUs/TPUs, iterating weekly. Edge case mining prioritized anomalies, boosting robustness 30x for rares like jaywalkers.
Safety incidents prompted hardware upgrades (new sensors) and software veto layers. Regulatory hurdles involved DMV/NHTSA reporting. GM's 2024 decision cited capital-intensive robotaxis vs. scalable personal AVs like Super Cruise 2.0, now enhanced by Cruise's AI models for hands/eyes-off by 2028.[7][8]
We build a proof of concept for your problem for 5,000–8,000€. You get a tangible demo instead of slides with promises.
Founder & Partner
Reruption GmbH
Falkertstraße 2
70176 Stuttgart
Phone