SS-Replan

Online observation, estimatation, planning, and control for a Franka Panda Robot operating in NVIDIA SRL's simulated kitchen environment.

Installation

SS-Replan supports both Python 2 and Python 3.

• Install Git LFS
• $ pip install numpy scipy pybullet sklearn
• $ git lfs clone --branch master --recurse-submodules https://github.com/caelan/SS-Replan.git
• $ cd SS-Replan
• SS-Replan$ git lfs install
• SS-Replan$ ./pddlstream/FastDownward/build.py release64
• SS-Replan$ cd ss-pybullet/pybullet_tools/ikfast/franka_panda
• SS-Replan/ss-pybullet/pybullet_tools/ikfast/franka_panda$ ./setup.py

It's also possible to use TRAC-IK instead of IKFast; however it requires installing ROS ($ sudo apt install ros-kinetic-trac-ik).

PyBullet Examples

• SS-Replan$ git pull
• SS-Replan$ git submodule update --init --recursive
• SS-Replan$ ./run_pybullet.py [-h]

IsaacSim Examples

Executed using the IssacSim 3D robot simulation environment.

Real-World Examples

Citation

Caelan R. Garrett, Chris Paxton, Tomás Lozano-Pérez, Leslie P. Kaelbling, Dieter Fox. Online Replanning in Belief Space for Partially Observable Task and Motion Problems, IEEE International Conference on Robotics and Automation (ICRA), 2020.

Publications

• Online Replanning in Belief Space for Partially Observable Task and Motion Problems
• PDDLStream: Integrating Symbolic Planners and Blackbox Samplers via Optimistic Adaptive Planning

Videos

• SS-Replan
• PDDLStream

Resources

• This repository uses PDDLStream to perform hybrid robotic planning.
• PDDLStream leverages FastDownward, a classical planner, as a discrete search subroutine.
• Common robotics primitives are implemented using PyBullet.

PDDLStream

• PDDLStream Paper
• PDDLStream Github Repository
• PDDLStream Tutorial

PDDL and FastDownward

• Planning Domain Definition Language (PDDL)
• PDDL Derived Predicates
• FastDownward Homepage

PyBullet

• PyBullet Package
• PyBullet Quickstart Guide