Autolife Planning

A planning library for the Autolife robot — inverse kinematics (TRAC-IK and Pink QP), motion planning (OMPL frontend, VAMP SIMD collision backend), and time-optimal trajectory generation, all behind a unified Python API.

• :material-package-variant:{ .lg .middle } Drop-in, few deps

  ---

  For inference: numpy, scipy, pink. No conda required, no ROS, no MoveIt. Vendor it into any project with a single pip install -e ..

• :material-source-branch:{ .lg .middle } 26 planners

  ---

  Full OMPL planner library, not a bespoke subset: RRT-Connect, RRT*, BIT*, AIT*, EIT*, KPIECE, PRM, FMT, SPARS, … plug in whichever suits the problem.

• :material-speedometer:{ .lg .middle } Microsecond checks, millisecond plans

  ---

  VAMP SIMD collision backend: ~3 μs per check. RRT-Connect plans a 1.1 ms collision-free path for the 7-, 14-, and 24-DOF body.

Features

• :material-robot-angry-outline:{ .lg .middle } Inverse Kinematics

  ---

  TRAC-IK for unconstrained numerical IK; Pink for QP-based constrained IK that composes end-effector tracking with centre-of-mass stability, camera-frame stabilization, and self-collision avoidance.

• :material-map-marker-path:{ .lg .middle } Motion Planning

  ---

  OMPL frontend + VAMP SIMD backend. Full-body and subgroup planning (single-arm, dual-arm, torso + arm, base, whole-body) with first-class support for point-cloud obstacles.

• :material-vector-curve:{ .lg .middle } Manifold Planning

  ---

  Hard task-space equality constraints — planes, rails, orientation locks, couplings — compiled from CasADi expressions and solved on OMPL's ProjectedStateSpace.

• :material-chart-bell-curve:{ .lg .middle } Cost-space Planning

  ---

  Path-integral soft costs (orientation preference, pose stabilization, …) compiled from CasADi and driving OMPL's asymptotically-optimal planners (RRT*, BIT*, AIT*, …).

• :material-timer-outline:{ .lg .middle } Time Parameterization

  ---

  Time-optimal trajectory generation (TOTG) converts geometric paths into executable trajectories with per-joint velocity and acceleration limits — bridging the planner's output to hardware.

Benchmarks

Median / p95 wall-clock for a single search (simplification and interpolation disabled), N = 20 runs per cell, plans from the home pose to a sampled collision-free goal around a tabletop point-cloud obstacle (same scene as examples/planning/motion.py).

Planner	Left arm (7 DOF)	Dual arm (14 DOF)	Full body (24 DOF)
RRT-Connect	1.1 / 1.1 ms	1.1 / 1.2 ms	1.1 / 1.1 ms
KPIECE	1.1 / 2.3 ms	1.7 / 3.2 ms	1.1 / 4.3 ms
PRM	1.1 / 3.2 ms	1.1 / 3.2 ms	2.2 / 4.2 ms
SBL	1.1 / 1.1 ms	2.2 / 3.3 ms	8.5 / 17.0 ms
EST / BiEST	1.1 / 4.3 ms	2.1 / 5.3 ms	2.2 / 6.4 ms
STRIDE	2.2 / 3.2 ms	2.7 / 8.5 ms	2.2 / 3.3 ms

Raw planner.validate(config) throughput: 3.2 μs / check (Python boundary included) ≈ 0.31 M checks / s. Batched planner.validate_batch(configs) packs up to rake distinct configs per SIMD block for an ~8.5× best-case speedup on all-valid workloads.

Environment

13th Gen Intel Core i9-13900KF, 24 logical cores · Linux 6.8 · Python 3.12.13. Reproduce with pixi run -e dev python scripts/benchmarks/motion_planning.py. See planning for the full planner sweep including asymptotically-optimal variants (RRT*, BIT*, AIT*, EIT*, FMT).

Quick Install

For inference — running the planners and IK solvers — just pip install:

git clone --recursive https://github.com/AdaCompNUS/Autolife-Planning.git
cd Autolife-Planning
pip install -e .

Three runtime deps: numpy, scipy, pink. No conda, no ROS, no MoveIt. See the Getting Started guide for the full development setup (pixi + conda-forge toolchain, URDF rebuilds, FK codegen).