Transformer Transformer: A Unified Model for Motion-Conditioned Robot Co-design

• URL: http://arxiv.org/abs/2607.25798v1
• Date: Tue, 28 Jul 2026 14:47:14 GMT
• Title: Transformer Transformer: A Unified Model for Motion-Conditioned Robot Co-design
• Abstract summary: Transformer Transformer is a dynamics model, whose reward-agnostic state and action predictions can be converted into reward-specific value predictions.<n>We fabricated an optimized ALOHA design, which reduced tracking error by over 70% compared to the original design.
• Score: 29.19489133790808
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: An often overlooked factor of robot manipulation performance is the embodiment of the robot itself. Motivated by this problem, we study motion-conditioned robot co-design, where the goal is to generate complete robot designs that track target end-effector trajectories (from human demonstrations) while optimizing user-defined rewards. We introduce Transformer Transformer, a diffusion transformer trained on RoboTokens, a unified tokenization of robot embodiments, states, and actions. The same architecture can be used across embodiment spaces (e.g., wheeled bimanual, quadrupeds, humanoids) and use cases (embodiment generation, cross embodiment controller). Rather than overfitting to one reward function, Transformer Transformer is a dynamics model, whose reward-agnostic state and action predictions can be converted into reward-specific value predictions. These value predictions are used to steer embodiment diffusion towards high value robot designs, through a procedure we call Dynamics Self-Guidance. Experiments across multiple design spaces show zero-shot optimization of unseen rewards and trajectories, improving performance and runtime over the evolutionary baseline. Finally, we fabricated an optimized ALOHA design, which reduced tracking error by over 70% compared to the original design.

Related papers

• Shape Your Body: Value Gradients for Multi-Embodiment Robot Design [16.640420524594443]
  We propose to turn generalist multi-embodiment value functions into reusable models for robot design.<n>Instead of running a new reinforcement learning co-design loop for each robot, we first train an embodiment-aware policy and value function across many robot designs.
  arXiv  Detail & Related papers  (2026-05-30T12:21:09Z)
• Multimodal embodiment-aware navigation transformer [5.673561896367136]
  ViLiNT is a multimodal, attention-based policy for goal navigation, trained on heterogeneous data from multiple platforms and environments.<n>We show that ViLiNT improves Success Rate on average by 166% over equivalent state-of-the-art vision-only baseline (NoMaD)<n>Results are confirmed through real-world deployments of a rover navigating in obstacle fields.
  arXiv  Detail & Related papers  (2026-04-21T09:32:59Z)
• Robometer: Scaling General-Purpose Robotic Reward Models via Trajectory Comparisons [69.87766750714945]
  General-purpose robot reward models are typically trained to predict absolute task progress from expert demonstrations.<n>We introduce Robometer, a scalable reward modeling framework that combines intra-trajectory progress supervision with inter-trajectory preference supervision.<n>Robometer is trained with a dual objective: a frame-level progress loss that anchors reward magnitude on expert data, and a trajectory-comparison preference loss that imposes global ordering constraints.
  arXiv  Detail & Related papers  (2026-03-02T17:38:58Z)
• RobotSeg: A Model and Dataset for Segmenting Robots in Image and Video [56.9581053843815]
  We introduce RobotSeg, a foundation model for robot segmentation in image and video.<n>It addresses the lack of adaptation to articulated robots, reliance on manual prompts, and the need for per-frame training mask annotations.<n>It achieves state-of-the-art performance on both images and videos.
  arXiv  Detail & Related papers  (2025-11-28T07:51:02Z)
• Physical Autoregressive Model for Robotic Manipulation without Action Pretraining [65.8971623698511]
  We build upon autoregressive video generation models to propose a Physical Autoregressive Model (PAR)<n>PAR leverages the world knowledge embedded in video pretraining to understand physical dynamics without requiring action pretraining.<n>Experiments on the ManiSkill benchmark show that PAR achieves a 100% success rate on the PushCube task.
  arXiv  Detail & Related papers  (2025-08-13T13:54:51Z)
• DiffGen: Robot Demonstration Generation via Differentiable Physics Simulation, Differentiable Rendering, and Vision-Language Model [72.66465487508556]
  DiffGen is a novel framework that integrates differentiable physics simulation, differentiable rendering, and a vision-language model. It can generate realistic robot demonstrations by minimizing the distance between the embedding of the language instruction and the embedding of the simulated observation. Experiments demonstrate that with DiffGen, we could efficiently and effectively generate robot data with minimal human effort or training time.
  arXiv  Detail & Related papers  (2024-05-12T15:38:17Z)
• Innate Motivation for Robot Swarms by Minimizing Surprise: From Simple Simulations to Real-World Experiments [6.21540494241516]
  Large-scale mobile multi-robot systems can be beneficial over monolithic robots because of higher potential for robustness and scalability. Developing controllers for multi-robot systems is challenging because the multitude of interactions is hard to anticipate and difficult to model. Innate motivation tries to avoid the specific formulation of rewards and work instead with different drivers, such as curiosity. A unique advantage of the swarm robot case is that swarm members populate the robot's environment and can trigger more active behaviors in a self-referential loop.
  arXiv  Detail & Related papers  (2024-05-04T06:25:58Z)
• Robot Learning with Sensorimotor Pre-training [98.7755895548928]
  We present a self-supervised sensorimotor pre-training approach for robotics. Our model, called RPT, is a Transformer that operates on sequences of sensorimotor tokens. We find that sensorimotor pre-training consistently outperforms training from scratch, has favorable scaling properties, and enables transfer across different tasks, environments, and robots.
  arXiv  Detail & Related papers  (2023-06-16T17:58:10Z)
• Efficient automatic design of robots [43.968830087704035]
  We show for the first time de-novo optimization of a robot's structure to exhibit a desired behavior, within seconds on a single consumer-grade computer. Unlike other gradient-based robot design methods, this algorithm does not presuppose any particular anatomical form. This advance promises near instantaneous design, manufacture, and deployment of unique and useful machines for medical, environmental, vehicular, and space-based tasks.
  arXiv  Detail & Related papers  (2023-06-05T21:30:52Z)
• REvolveR: Continuous Evolutionary Models for Robot-to-robot Policy Transfer [57.045140028275036]
  We consider the problem of transferring a policy across two different robots with significantly different parameters such as kinematics and morphology. Existing approaches that train a new policy by matching the action or state transition distribution, including imitation learning methods, fail due to optimal action and/or state distribution being mismatched in different robots. We propose a novel method named $REvolveR$ of using continuous evolutionary models for robotic policy transfer implemented in a physics simulator.
  arXiv  Detail & Related papers  (2022-02-10T18:50:25Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.