Skip to content
RobothecaBeta
Log in

pytorch_kinematics

  • Software
  • Robot arms and industry
  • Library

A PyTorch library for differentiable, batched robot forward kinematics, Jacobians and damped least squares inverse kinematics, loading URDF, SDF and MJCF robot descriptions.

Made by The Autonomous Robotic Manipulation Lab.

Source on GitHub: github.com/UM-ARM-Lab/pytorch_kinematics (opens in a new tab). Facts read at commit 295e225 (opens in a new tab).

Licenses

Code
MIT (opens in a new tab)

Read by Robotheca from LICENSE.txt (opens in a new tab) (MIT) at commit 295e225 (opens in a new tab), on 16 Oct 2026.

Commercial use
Yes

What it does

  • Loads robot descriptions from URDF, SDF and MJCF files and builds kinematic trees or serial chains.
  • Computes forward kinematics in parallel over batches of joint configurations, with gradients through the result.
  • Provides an analytical backward pass for forward kinematics that is faster than standard autograd on GPU.
  • Computes differentiable, parallelizable Jacobians for serial chains, optionally at offset points on the end effector.
  • Solves inverse kinematics with damped least squares, batched over goals and random retries, including goal orientation.
  • Offers torch.compile-compatible variants of forward kinematics, Jacobian and inverse kinematics for faster repeated calls.

Written by Robotheca from the project's README (opens in a new tab) at commit 295e225.

Getting started

Difficulty
Intermediate
Hardware it needs, as the project states it
CUDA GPU is optional and used when available; otherwise runs on CPU.

Maintenance

Status
Maintained: a commit in the last 12 months
Newest commit
16 Apr 2026 on master
Latest release
v0.10.0 (16 Apr 2026)

Checked on GitHub on 16 Oct 2026.

Links

Robots that can run it

  • Robot arm

    U-Arm

    A low-cost, universal leader-follower teleoperation system for commercial robot arms with plug-and-play hardware configurations.

    • Works with AI
    Level
    Intermediate
    Buildability
    75/100
    Buildable with some effort
    Cost
    $60≈ €53 (ECB rate, 16 Oct 2026)

    Found by Robotheca in requirements.txt (opens in a new tab)

Robotheca lists software as a reference: nothing is hosted or run here, and every fact links to where it was read.