Primary question: Does your vehicle appear on the 300+ supported list, and are you comfortable running community-developed driver assistance software that ships with no warranty?
RepoDaily adoption score
RepoDaily rates this as 85/100 (strong) for adoption: evidence, installation path, production risk, differentiation, license clarity, and AI/agent fit are scored from the article sources and adoption notes.
6 source(s) across 4 source category/categories, plus a RepoDaily-specific evidence module when available.
6 workflow step(s), 5 next-action step(s), and 4 command/install signal(s) were detected.
Trending momentum is +465 stars, with maintenance/release/issue signals counted when present.
Risk is marked high, with 6 security note(s) and 4 explicit skip condition(s).
3 opportunity lens item(s), 4 alternative(s), and 3 type-specific section(s) support differentiation.
License source or license wording is present.
4 AI/agent-related signal(s) were detected in the article text and metadata.
Project overview
openpilot is an operating system for robotics built by Comma.ai. Its primary use today is upgrading the factory driver assistance system on over 300 supported cars — adding enhanced lane-keeping and adaptive cruise control through a comma device mounted on the windshield. The project is written primarily in Python, ships under the MIT License, and has been in active development since at least 2018.
The latest releases show a project in deep technical evolution. Version 0.10.0 (August 2025) introduced a new driving model architecture described in Comma's CVPR paper 'Learning to Drive from a World Model,' replacing the longitudinal MPC with end-to-end planning in Experimental Mode. By version 0.10.2 (November 2025), openpilot added support for the new comma four hardware. Version 0.11.0 (March 2026) brought a model fully trained using a learned simulator and cut comma four standby power usage by 77% to 52 mW. The most recent release, 0.11.1 (May 2026), shipped a new driver monitoring model and added support for the Acura MDX 2022-24 and Rivian R1S/R1T 2025.
From a security and safety perspective, openpilot occupies unusual ground. It is open-source software that directly controls vehicle steering and acceleration, yet it ships under an MIT License that explicitly disclaims all warranties. The project maintains a SECURITY.md policy directing vulnerability reports to adeeb@comma.ai and security@comma.ai. Driver monitoring — a camera-based system that tracks head pose and detects distraction — has been iterated in nearly every recent release, serving as a primary safety guardrail.
Why it is trending now
- Version 0.11.1 shipped on 2026-05-18 with a new driver monitoring model and improved thermal policy for comma four, plus Acura MDX 2022-24 and Rivian R1S/R1T 2025 support
- The driving model architecture introduced in 0.10.0 uses end-to-end planning from a World Model, a approach described in a CVPR paper — a notable shift from traditional MPC-based control
- comma four hardware support (0.10.2, November 2025) brought a new device generation, and 0.11.0 cut its standby power by 77% to 52 mW
- Support now spans 300+ vehicles across Toyota, Honda, Ford, Hyundai, Kia, Lexus, Tesla, Rivian, and others, with community contributors regularly adding new models
Problem it solves
- Factory ADAS systems on many vehicles lack robust lane-keeping or have limited adaptive cruise behavior — openpilot augments or replaces these with its own models
- Traditional MPC-based longitudinal control can feel jerky or inconsistent; the World Model approach in Experimental Mode replaces it with learned end-to-end planning
- Driver distraction is a leading cause of accidents; openpilot's driver monitoring model watches head pose and flags distraction, with continuous improvements to reduce false positives from passengers
- Power draw on always-on devices matters for battery life; the 77% standby power reduction on comma four addresses this directly
How it works
- A comma device (comma 3X or comma four) is mounted on the windshield and connected to the vehicle's CAN bus, typically through the OBD-II port or a custom harness
- The device runs AGNOS, Comma's Linux-based operating system, which hosts the openpilot software stack including the driving model, vision pipeline, and driver monitoring
- Camera input feeds a neural network driving model that outputs desired curvature for lateral control and, in Experimental Mode with the World Model, longitudinal planning
- A separate driver monitoring camera tracks the driver's head pose and eye attention, disengaging or alerting when distraction is detected
- Vehicle fingerprinting identifies the car make, model, and year — since 0.9.7 this works without the OBD-II port on all cars, using improved fuzzy fingerprinting
- Control commands are sent back through the CAN bus to the vehicle's factory steering and throttle actuators within the supported platform's limits
Integration Surface: Dependencies, Packages, and Python Requirements
- Requires Python >= 3.12.3, < 3.13 per pyproject.toml — a narrow version window
- Build system is hatchling; wheel packages include openpilot, msgq, opendbc, panda, rednose, teleoprtc, and tinygrad
- Core dependencies include pycapnp==2.1.0 (pinned because 2.2 introduces a memory leak from cyclic references), numpy >=2.0, pyzmq, sentry-sdk, Cython, and cffi
- Vendored native dependencies pulled from commaai/dependencies.git include capnproto, ffmpeg, libyuv, zeromq, eigen, acados, raylib, and zstd
- The project uses uv with python-preference = 'only-managed', meaning it manages its own Python installation rather than relying on system Python
- Ruff is configured with line-length 160, indent-width 2, and bans specific APIs like pytest.main, unittest, and several raylib functions via flake8-tidy-imports
Maintenance Risk: Release Cadence and Update Patterns
- Releases are frequent: 0.9.8 through 0.11.2 span February 2025 to June 2026, roughly every 1-3 months
- Each release typically includes a new driving model, often a new driver monitoring model, and new vehicle support — indicating active, sustained development
- Version 0.11.2 (2026-06-15) is listed but has no changelog entries yet, suggesting it was published very recently relative to the source pack
- Community contributors are named in release notes (e.g., sunnyhaibin for Hyundai models, lukasloetkolben for Rivian and Tesla), showing distributed vehicle-support contributions
- The localizer was rewritten in 0.9.8 to remove GPS dependency at runtime — a significant architectural change that could affect users who relied on the old behavior
Deployment Notes: Hardware and Platform Requirements
- comma four is the current generation device (supported since 0.10.2, November 2025); comma 3X was supported since 0.9.4 (July 2023)
- The software runs on AGNOS, Comma's custom Linux image — not a generic desktop or server OS
- Python must be 3.12.x (>= 3.12.3, < 3.13); the project uses uv to manage this automatically
- pytest markers indicate some tests are C3/C3X-specific (marked 'tici'), meaning the full test suite cannot run on all hardware
- Fingerprinting without OBD-II has been available since 0.9.7 (June 2024), simplifying installation on supported vehicles
Who should pay attention?
Good fit if
- Your vehicle is on the 300+ supported list and you already own or plan to purchase a comma 3X or comma four device
- You want to contribute vehicle support for an unsupported model — the release notes regularly credit community contributors
- You are a robotics researcher interested in the World Model architecture or learned-simulator training approach described in the CVPR paper
- You want to audit or study a production-grade open-source ADAS stack that controls real vehicles
Skip for now if
- You drive a vehicle not on the supported list — openpilot cannot be safely installed without community porting work
- You expect warranty coverage or liability protection — the MIT License explicitly disclaims all warranties for this safety-critical software
- You want a fully autonomous driving system — openpilot is a Level 2 driver assistance system that requires constant driver supervision
- You are looking for a software-only solution without dedicated hardware — openpilot requires a comma device and vehicle harness
Risks and cautions
openpilot controls steering and acceleration on real vehicles under a no-warranty MIT License. Adoption requires compatible hardware, a supported car, and acceptance of safety-critical software maintained by a community plus a private company.
- The MIT License states 'THE SOFTWARE IS PROVIDED AS IS, WITHOUT WARRANTY OF ANY KIND' — no fitness for purpose guarantees for software that controls vehicle dynamics
- The driving model is iterated frequently (new models in nearly every release), meaning behavior can change between updates
- Experimental Mode uses end-to-end planning from the World Model, replacing the traditional longitudinal MPC — a fundamentally different control approach
- Installation requires physical hardware (comma device), vehicle-specific harnesses, and CAN bus access
- The supported vehicle list changes with each release; some features like longitudinal control are alpha or experimental for certain makes
- SECURITY.md specifies that suspected vulnerabilities should be reported to both adeeb@comma.ai and security@comma.ai — a dual-channel reporting policy
- The MIT License explicitly disclaims all warranties including fitness for a particular purpose, which is significant for safety-critical automotive software
- Driver monitoring is a core safety feature: the camera-based system tracks head pose and distraction, and has received model updates in versions 0.9.8, 0.10.1, 0.10.3, 0.11.1, and others
- Since 0.9.8, openpilot can fingerprint vehicles without the OBD-II port, reducing the need for persistent CAN bus probing
- sentry-sdk is a declared dependency, meaning the software may send error/crash telemetry to Sentry infrastructure
- pycryptodome is listed as a dependency used in panda, body, and a test, indicating cryptographic operations in the CAN bus interface layer
Alternatives to compare
| Approach | When to use | Trade-off |
|---|---|---|
Tesla Autopilot / Full Self-Driving | You own a Tesla — the factory system is tightly integrated and openpilot is not designed to replace it | Included or subscription-based, depending on Tesla's current pricing |
Autoware | You are building autonomous driving research prototypes rather than upgrading a consumer vehicle's ADAS | Open source (Apache 2.0) |
Apollo Auto | You need a full autonomous driving stack for research or commercial fleet development | Open source (Apache 2.0) |
Factory ADAS (vehicle built-in) | You want full manufacturer warranty coverage and liability protection — no aftermarket modification | Included with vehicle purchase |
What this trend reveals
Vehicle Support Contributions
Release notes regularly credit community contributors for adding new vehicle support (e.g., mvl-boston for Acura MDX, lukasloetkolben for Rivian and Tesla). If you own a vehicle not yet supported, contributing CAN bus fingerprinting data and testing could add it.
Check the 300+ supported vehicles list at comma.ai/openpilot and review the community contributions pattern in RELEASES.md
World Model Research Replication
The 0.10.0 release references a CVPR paper titled 'Learning to Drive from a World Model.' Researchers can study how the temporal policy architecture and on-policy training physics noise model evolved across 0.10.0, 0.10.1, 0.10.3, and 0.11.0.
Read the CVPR paper and compare the model architecture changes documented across four consecutive release notes
Power Optimization on Embedded Hardware
The 77% standby power reduction on comma four (to 52 mW in 0.11.0) and the 0.5W GPU power reduction from ISP image processing (0.9.8) demonstrate concrete techniques for embedded robotics power management.
Review the 0.11.0 and 0.9.8 release notes for specific power metrics and the architectural changes that enabled them
RepoDaily verdict
openpilot is the most mature open-source driver assistance system available, with a rapid release cadence, a research-grade World Model architecture, and 300+ supported vehicles. Its MIT License and no-warranty stance make it a serious adoption decision for safety-critical automotive use — but for developers, researchers, and informed vehicle owners, it is an exceptional technical project to study, contribute to, or deploy with appropriate caution.