# andyzeng/visual-pushing-grasping

Train robotic agents to learn to plan pushing and grasping actions for manipulation with deep reinforcement learning.

Repository: https://github.com/andyzeng/visual-pushing-grasping
Canonical: https://ross.abutalabs.com/products/visual-pushing-grasping
Homepage: http://vpg.cs.princeton.edu/
Language: Python
License: BSD-2-Clause
License Family: permissive
Topics: robotics, manipulation, grasping, pushing, computer-vision, deep-learning, 3d, vision, deep-reinforcement-learning, artificial-intelligence
Last push: 2021-05-11T14:06:29+00:00

## Health v2 (maintenance only)
Score: 32/100 (v2, computed 2026-09-03T02:20:16.233290+00:00)
- activity 0, release rhythm 35, longevity 100
- inputs: {"age_days": 3073, "days_push": 1940, "days_rel": null, "gap_med": null, "n_releases_24m": 0}
- flags: no_releases
- formula: round(0.45*activity + 0.35*rhythm + 0.20*longevity); archived -> min(score, 10)

## Adoption (not part of the score)
Stars 1109, forks 328 (observed 2026-08-28T04:03:37.374964+00:00)

## What it is
PyTorch reference implementation of Visual Pushing and Grasping (VPG), which trains robotic agents via self-supervised deep reinforcement learning to plan complementary pushing and grasping actions from RGB-D observations. It supports training and testing in simulation and on a real UR5 robot arm, based on the IROS 2018 paper.

## Use cases
- train a robot to grasp objects in cluttered scenes
- learn pushing and grasping policies with deep reinforcement learning
- run robotic manipulation experiments in simulation and on a real UR5 arm
- reproduce results from the visual pushing and grasping paper
- train self-supervised manipulation policies from RGB-D images
- research synergies between non-prehensile and prehensile robot actions

## When to choose
- you need a proven reference implementation of deep RL for robotic pushing and grasping
- you want to train manipulation policies from RGB-D visual observations with self-supervision
- you are doing research on robotic manipulation and want a well-cited baseline

## When to avoid
- you need a production-ready or actively maintained robotics framework
- you use robot hardware other than a UR5 arm without adaptation work
- you need general-purpose robot simulation tooling rather than this specific VPG method

## Facets
- artifact type: library
- maturity: maintenance
- function: machine-learning, deep-learning, reinforcement-learning, computer-vision, robotics
- domain: robotics, reinforcement-learning, computer-vision, artificial-intelligence
- platform: python
- tags: robot-manipulation, grasping, pushing, pytorch, q-learning, self-supervised-learning, rgb-d, research-code, ur5-robot-arm, linux, gpu

## Member repositories
- andyzeng/visual-pushing-grasping (main) score 32

## Provenance
- Observed fields: from GitHub, fetched 2026-08-28T04:03:37.374964+00:00.
- Health v2: computed from the inputs above; adoption is never an input.
- Inferred fields (summary, facets, guidance): AI-extracted, prompt v1, taxonomy v1, on 2026-08-30T06:43:33.543786+00:00, confidence not recorded.
  - readme: https://github.com/andyzeng/visual-pushing-grasping (fetched 2026-08-28T04:03:37.374964+00:00, sha eff9e02c1cec)
  - homepage: http://vpg.cs.princeton.edu/ (fetched 2026-08-29T12:47:26.787223+00:00, sha ec04e6403aae)
- Data as of 2026-08-30T08:39:29.467469+00:00.