[Amazon Warehouse] Study of new version of OMPL, 2.0

[jmplaza]

Currently we are using OMPL 1.7.0 in the Amazon Warehouse exercise. Presentation of the new release at the official OMPL GitHub: https://github.com/ompl/ompl/releases/tag/2.0.0

The new release includes several additions, including VAMP integration, and changes aimed at improving functionality, compatibility, and the user experience. Below is a summary of the main updates:

Completely rewritten Python bindings.
Enabled by default if Python is detected at build time.
Prebuilt OMPL Python modules can now be installed via pip install ompl.
VAMP Integration
Added VAMP (Vector-Accelerated Motion Planning) as an optional high-performance backend for collision checking and motion validation
VAMP leverages SIMD instructions to accelerate forward kinematics and collision detection, achieving planning speeds up to 25 kHz
C++ and Python demos show usage with common robots (Panda, UR5, Fetch, Baxter)
New geometric planners:
AORRTC: Asymptotically Optimal RRT-Connect
BLIT*: Bidirectional Lazy Informed Trees
TRRT*, ATRRT: asymptotically optimal version of T-RRT and an anytime (optimal) version of T-RRT, respectively
New kinodynamic planner: HySST, an adaptation of the SST planner for hybrid systems.
New state space: ThrochoidStateSpace, an SE(2)-like state space where distance and interpolation is defined for Dubins vehicles subject to constant drift. This is useful in planning for aerial/maritime drones subject to constant wind/current.
Planner Arena has been completely rewritten in Python. It is now distributed separately and can be installed via pip install plannerarena.
OMPL.app has been deprecated. New demos show how to use OMPL with real robots and visualize the results, eliminating the need for OMPL.app.
Bug fixes.

Download

You can download OMPL and find more information about OMPL on its homepage at https://ompl.kavrakilab.org.

[Sarvesh-Mishra1981]

Hi @jmplaza sir, I was just looking into this. Back in PR #3409 , I tried upgrading to OMPL 1.7.0, but the python bindings were pretty unstable in our Docker environment and gave us some headaches.

The best part about 2.0.0 is that they completely rewrote those python bindings. This time we can ditch our old installation script entirely and just use pip install ompl which should easily fix the old instability issues we had.

Also, for the Amazon Warehouse exercise, standard collision checking can get slow in tight spaces. Since 2.0.0 adds the new VAMP backend I think it would be really interesting to benchmark it and see if it speeds up the robot navigation in the aisles.

Let me know what you think If this sounds like a good plan I can start doing some local tests.

[Kunal-Somani]

Hi @jmplaza, here's what I found by testing inside the current RADI container:

Current state:

• C++ library: OMPL 1.7.0 (installed at /usr/local/include/ompl-1.7/)
• Python bindings: partially updated — located at /usr/lib/python3/dist-packages/ompl and expose OMPL 2.0 state spaces (VanaStateSpace, OwenStateSpace, STATE_SPACE_VANA) but are backed by the 1.7 C++ library

What works:

• Core planning: SE2StateSpace, RRTstar, SpaceInformation all functional
• OMPL 2.0 state spaces present in Python bindings: VanaState, OwenState, VanaOwenState

What does NOT work:

• New OMPL 2.0 planners (AORRTC, BLITstar, TRRTstar, ATRRT) are all missing — not compiled into the C++ backend
• pip install ompl is not used — OMPL is installed from source

Conclusion:
The current RADI has a partial 2.0 migration in which Python bindings expose some 2.0 API but the C++ library is still 1.7. A full upgrade would require rebuilding OMPL 2.0 from source in the RADI Dockerfile. The existing Amazon Warehouse exercise code runs correctly against the current setup.

Shall I open a PR upgrading the RADI Dockerfile to build OMPL 2.0 from source?