carvalho_robot.png

João Carvalho

Postdoctoral Researcher @ TU Darmstadt

contact: joao@robot-learning.de

google scholar • github • linkedin • twitter • bluesky • ias webpage

I am a Postdoctoral Researcher at the Intelligent Autonomous Systems (IAS) group from TU Darmstadt, where I recently obtained my Ph.D. in Computer Science advised by Jan Peters.
My research interests are developing machine learning and reinforcement learning algorithms for robot manipulation. I've developed algorithms that use deep generative models for visuomotor policies, motion planning and grasping, and on reinforcement learning methods to solve contact-rich tasks like insertions. In the past, I've also worked on sample efficient off-policy reinforcement learning algorithms, and on variance reduction techniques for policy gradients.


news

13 Mar 2025 Four papers accepted at the 1st German Robotics Conference.
01 Feb 2025 I started a new position as Postdoctoral Researcher at IAS.
17 Jan 2025 I’ve defended my Ph.D. thesis on Enhancing Robot Manipulation Skills through Learning.

key publications

For a complete and up-to-date list, please visit my google scholar.

2025

  1. mpd-splines.png
    Motion Planning Diffusion: Learning and Adapting Robot Motion Planning With Diffusion Models
    João Carvalho, An T. Le, Piotr Kicki, Dorothea Koert, and Jan Peters
    IEEE Trans. Robotics, 2025
    PDF Website Code
  2. cube_mtp-akima.png
    Model Tensor Planning
    An Thai Le, Khai Nguyen, Minh Nhat Vu, João Carvalho, and Jan Peters
    Trans. Mach. Learn. Res., 2025
    PDF Website Code
  3. gtmp_spline_occupancy.png
    Global Tensor Motion Planning
    An T. Le, Kay Pompetzki, João Carvalho, Joe Watson, Julen Urain, and 3 more authors
    IEEE Robotics Autom. Lett., 2025
    PDF Website Code

2024

  1. gdn.png
    Grasp Diffusion Network: Learning Grasp Generators from Partial Point Clouds with Diffusion Models in SO(3)xR3
    João Carvalho, An T. Le, Philipp Jahr, Qiao Sun, Julen Urain, and 2 more authors
    CoRR, 2024
    PDF Website
  2. actionflow.png
    ActionFlow: Equivariant, Accurate, and Efficient Policies with Spatially Symmetric Flow Matching
    Niklas Funk, Julen Urain, Joao Carvalho, Vignesh Prasad, Georgia Chalvatzaki, and 1 more author
    CoRR, 2024
    PDF Website

2023

  1. mpd.png
    Motion Planning Diffusion: Learning and Planning of Robot Motions with Diffusion Models
    João Carvalho, An T. Le, Mark Baierl, Dorothea Koert, and Jan Peters
    In IEEE/RSJ International Conference on Intelligent Robots and Systems IROS, 2023
    PDF Website Code

2022

  1. rrl_promp.png
    Adapting Object-Centric Probabilistic Movement Primitives with Residual Reinforcement Learning
    João Carvalho, Dorothea Koert, Marek Daniv, and Jan Peters
    In 21st IEEE-RAS International Conference on Humanoid Robots, Humanoids 2022, Ginowan, Japan, November 28-30, 2022, 2022
    PDF
  2. conditioned_sbm_trajectory.png
    Conditioned Score-Based Models for Learning Collision-Free Trajectory Generation
    João Carvalho, Mark Baierl, Julen Urain, and Jan Peters
    In NeurIPS 2022 Workshop on Score-Based Methods, 2022
    PDF
  3. nopg_mountain_car.png
    Batch Reinforcement Learning With a Nonparametric Off-Policy Policy Gradient
    Samuele Tosatto, João Carvalho, and Jan Peters
    IEEE Trans. Pattern Anal. Mach. Intell., 2022
    PDF Code

2021

  1. mvd-rl.png
    An Empirical Analysis of Measure-Valued Derivatives for Policy Gradients
    João Carvalho, Davide Tateo, Fabio Muratore, and Jan Peters
    In International Joint Conference on Neural Networks, IJCNN 2021, Shenzhen, China, July 18-22, 2021, 2021
    PDF Code

2020

  1. nopg_lqr.png
    A Nonparametric Off-Policy Policy Gradient
    Samuele Tosatto, João Carvalho, Hany Abdulsamad, and Jan Peters
    In The 23rd International Conference on Artificial Intelligence and Statistics, AISTATS 2020, 26-28 August 2020, Online [Palermo, Sicily, Italy], 2020
    PDF Code