HN Debrief

Box3D, an open source 3D physics engine

  • Open Source
  • Game Development
  • Programming
  • Developer Tools

The post announces Box3D, a new open source 3D physics engine from Erin Catto, best known for Box2D. The announcement frames it as a rigid-body engine built for games, with demo footage and documentation that show collision handling, replay support, and integrations beyond a toy prototype. That mattered because Box2D has a long track record in games and simulation tooling, from indie titles to reinforcement learning benchmarks like OpenAI Gym's LunarLander and CarRacing, so people came in expecting practical code rather than a research project.

If you ship games, browser apps, or simulations and have been defaulting to Bullet, PhysX, Jolt, or engine-built-ins, Box3D is worth immediate evaluation for projects that need reproducible physics and a lightweight integration path. The bigger strategic signal is that open source game infrastructure is still under-supplied in 3D, so a credible new core engine can quickly become ecosystem-defining.

Discussion mood

Strongly positive and a little nostalgic. People trust Erin Catto's track record, like that Box3D appears small and practical, and see real value in a new open source 3D engine with determinism. The only sustained frustration was about the old Box2D compensation and credit story, which reopened the usual anxiety about permissive open source economics.

Key insights

  1. 01

    Cross-platform determinism is already documented

    The docs appear to promise something far more valuable than generic replay support. Box3D is designed to produce deterministic results across both thread counts and platforms, while also rejecting compiler settings like -ffast-math that would undermine that guarantee. That puts it in a much narrower and more useful category for networked games than most engine physics stacks.

    If your roadmap includes rollback netcode, authoritative replays, or lockstep simulation, test Box3D early instead of assuming you will need custom physics. Verify the determinism claims on your actual target platforms before you commit.

      Attribution:
    • tikotus #1 #2
  2. 02

    Small binary and likely easy WASM path

    A release build coming in under 1 MB on macOS changed the tone from "interesting new engine" to "plausible for the web." The combination of small size, straightforward build, and a C API suggests Box3D may dodge the usual WebAssembly integration pain that makes many 3D physics engines unattractive for browser games.

    If you have been avoiding full 3D physics in browser projects because of payload size or toolchain friction, put Box3D on a benchmark list now. Compare startup cost, build complexity, and runtime footprint against JoltPhysics.js and PhysX-based options.

      Attribution:
    • tikotus #1
  3. 03

    Open source 3D physics is still a thin market

    The notable thing is not just that Box3D exists. It is that the credible open source shortlist is still short enough that people can rattle it off from memory. ODE, Bullet, Newton Dynamics, PhysX, Jolt, Rapier, and Avian cover most of the serious discussion. That scarcity helps explain the enthusiasm. A strong entrant can matter quickly because teams do not have dozens of mature alternatives.

    Treat physics engine choice as a strategic dependency, not a commodity checkbox. A small number of projects will shape portability, networking architecture, and tooling for years, so keep a close eye on which ecosystem is actually gaining momentum.

      Attribution:
    • flohofwoe #1
    • mikulas_florek #1
    • tapirl #1
    • zokier #1
    • dude250711 #1
  4. 04

    Box2D mattered beyond games

    Its role in standard reinforcement learning environments is a reminder that Box2D escaped the game niche long ago. Tools like LunarLander and CarRacing in Gymnasium turned it into part of the benchmark substrate for machine learning work, which helps explain why a Box3D release draws interest from people outside traditional game development.

    If you build simulation-heavy ML or robotics tooling, watch Box3D as a candidate substrate for richer 3D benchmark environments. A stable open source engine with deterministic behavior can be more useful than a flashier game-engine stack.

      Attribution:
    • ainch #1
  5. 05

    The license debate exposed a packaging tradeoff

    The sharpest point in the Box2D compensation argument was that permissive licensing is doing two jobs at once. MIT made Box2D simple to adopt and legally legible for small teams, but that same simplicity removed any built-in path to revenue sharing. Suggestions like dual licensing, as used by CGAL, were raised as a way to preserve access while giving creators leverage. The deeper point is that maintainers often optimize for adoption first, then regret the economics later.

    If you maintain foundational developer infrastructure, revisit your license before usage explodes. Once an ecosystem standard forms around permissive terms, changing the business model gets much harder.

      Attribution:
    • sosodev #1
    • guyomes #1
    • CooCooCaCha #1
    • mvdtnz #1

Against the grain

  1. 01

    Great for simulation, less ideal for game feel

    For some 2D game developers, Box2D's weakness was never correctness. It was that generalized rigid-body physics fights the handcrafted movement and collision exceptions that make arcade games feel good. The newer C rewrite may improve the API, but it does not solve the deeper complaint that many games need escape hatches, soft bodies, deformers, and custom character behavior more than faithful physics.

    Do not equate a strong physics engine with a strong gameplay foundation. If your game depends on heavily tuned movement, prototype the control layer first and confirm the engine will not force you into constant workarounds.

      Attribution:
    • minraws #1 #2
    • a1o #1
  2. 02

    MIT did exactly what MIT says

    The pushback on the Angry Birds fairness story was blunt. If a maintainer chooses MIT, broad uncompensated commercial use is the deal, and moral disappointment after the fact does not change that. Once the credit issue was corrected, the remaining complaint was about generosity, not license violation or debt.

    If compensation or usage conditions matter to you, encode them up front. Counting on goodwill from successful adopters is not a business model.

      Attribution:
    • sosodev #1 #2
    • SiempreViernes #1
    • maccard #1

In plain english

-ffast-math ↩
A compiler option that enables aggressive floating-point optimizations, often at the cost of strict numerical correctness and reproducibility.
Avian ↩
An open source physics engine for the Bevy ecosystem, previously known as Bevy XPBD.
Box2D ↩
A long-running open source 2D physics engine widely used in games and other simulations.
Box3D ↩
A new open source 3D rigid-body physics engine from the creator of Box2D.
Bullet ↩
A widely used open source 3D physics engine for collision detection and rigid-body simulation.
C API ↩
An application programming interface exposed in the C programming language, often valued because it is simple to call from many languages and platforms.
CGAL ↩
Computational Geometry Algorithms Library, a geometry library that uses dual open source and commercial licensing.
Emscripten ↩
A toolchain that compiles C and C++ code to WebAssembly and JavaScript for use on the web.
Gymnasium ↩
A maintained successor to OpenAI Gym that provides reinforcement learning environments and APIs.
Jolt ↩
Jolt Physics, a modern 3D physics engine used in games and also available with web-targeted builds.
ODE ↩
Open Dynamics Engine, an older open source 3D physics engine.
OpenAI Gym ↩
A popular software toolkit for reinforcement learning experiments that provides standard environments for training and benchmarking agents.
PhysX ↩
NVIDIA's physics engine for games and simulation, later released as open source.
Rapier ↩
A modern physics engine written in Rust that supports 2D and 3D simulation.
reinforcement learning ↩
A machine learning approach where an agent learns by taking actions and receiving rewards from an environment.

Reference links

Physics engine references

Box3D documentation and demos

Licensing and open source economics

Related projects and applications