---
_id: '7175'
abstract:
- lang: eng
  text: Deep reinforcement learning (RL) policies for robotic control typically overfit
    to a single hardware configuration. Any change to the kinematic chain breaks the
    learned mapping and requires complete retraining. This paper investigates whether
    combining self-attention mechanisms with explicit spatial observations can improve
    a policy’s adaptability to varying kinematic topologies. We train a single RL
    agent to control a robotic manipulator across different degrees of freedom (DOF),
    ranging from a restricted 4-DOF mode to a fully redundant 7-DOF configuration.
    Instead of fixed-length state vectors, the method processes the active joints
    as a variable-length sequence in an attention buffer, enriching each joint’s representation
    with its relative spatial routing and geometric Jacobian influence. This structure
    allows the agent to dynamically evaluate the physical utility of its available
    actuators. The proposed Architecture reaches a success rate of 82.1% across trained
    topologies and 66.6% on unseen configurations, outperforming MLP and generic attention
    baselines while remaining nearly collisionfree. Finally, we demonstrate successful
    sim-to-real transfer by deploying the simulation-trained agent on a physical Franka
    Emika Panda manipulator.
author:
- first_name: Patrick Thomas
  full_name: Mayer, Patrick Thomas
  id: '244831'
  last_name: Mayer
  orcid: 0009-0009-6800-1273
  orcid_put_code_url: https://api.orcid.org/v2.0/0009-0009-6800-1273/work/227375532
- first_name: Jan
  full_name: Rexilius, Jan
  id: '245736'
  last_name: Rexilius
  orcid: 0000-0002-4579-214X
  orcid_put_code_url: https://api.orcid.org/v2.0/0000-0002-4579-214X/work/227375533
citation:
  alphadin: '<span style="font-variant:small-caps;">Mayer, Patrick Thomas</span> ;
    <span style="font-variant:small-caps;">Rexilius, Jan</span>: Adapting to Variable
    Kinematic Configurations: A Causal-Kinematic Attention Approach for Robotic Control.
    In:'
  ama: 'Mayer PT, Rexilius J. Adapting to Variable Kinematic Configurations: A Causal-Kinematic
    Attention Approach for Robotic Control.'
  apa: 'Mayer, P. T., &#38; Rexilius, J. (n.d.). Adapting to Variable Kinematic Configurations:
    A Causal-Kinematic Attention Approach for Robotic Control. Presented at the 9th
    Iberian Robotics Conference (ROBOT) , Barcelona.'
  bibtex: '@inproceedings{Mayer_Rexilius, title={Adapting to Variable Kinematic Configurations:
    A Causal-Kinematic Attention Approach for Robotic Control}, author={Mayer, Patrick
    Thomas and Rexilius, Jan} }'
  chicago: 'Mayer, Patrick Thomas, and Jan Rexilius. “Adapting to Variable Kinematic
    Configurations: A Causal-Kinematic Attention Approach for Robotic Control,” n.d.'
  ieee: 'P. T. Mayer and J. Rexilius, “Adapting to Variable Kinematic Configurations:
    A Causal-Kinematic Attention Approach for Robotic Control,” presented at the 9th
    Iberian Robotics Conference (ROBOT) , Barcelona.'
  mla: 'Mayer, Patrick Thomas, and Jan Rexilius. <i>Adapting to Variable Kinematic
    Configurations: A Causal-Kinematic Attention Approach for Robotic Control</i>.'
  short: 'P.T. Mayer, J. Rexilius, in: n.d.'
conference:
  end_date: 2026-11-20
  location: Barcelona
  name: '9th Iberian Robotics Conference (ROBOT) '
  start_date: 2026-11-18
date_created: 2026-09-21T16:34:09Z
date_updated: 2026-09-21T19:31:42Z
department:
- _id: '102'
keyword:
- Reinforcement Learning
- Robotic Manipulation
- Morphology Generalization
- Sim-to-Real Transfer
language:
- iso: eng
project:
- _id: A827C0AA-C7DA-11E9-B0AE-1F4CB252D58D
  name: Institute for Building Intelligence
publication_status: accepted
quality_controlled: '1'
status: public
title: 'Adapting to Variable Kinematic Configurations: A Causal-Kinematic Attention
  Approach for Robotic Control'
type: conference
user_id: '244831'
year: '2026'
...