GeCCo - a Generalist Contact-Conditioned Policy for Loco-Manipulation Skills on Legged Robots

Abstract

Most modern approaches to quadruped locomotion focus on using Deep Reinforcement Learning (DRL) to learn policies from scratch, in an end-to-end manner. Such methods often fail to scale, as every new problem or application requires time-consuming and iterative reward definition and tuning.

We present Generalist Contact-Conditioned Policy (GeCCo) --- a low-level policy trained with Deep Reinforcement Learning that is capable of tracking arbitrary contact points on a quadruped robot. The strength of our approach is that it provides a general and modular low-level controller that can be reused for a wider range of high-level tasks, without the need to re-train new controllers from scratch. We demonstrate the scalability and robustness of our method by evaluating on a wide range of locomotion and manipulation tasks in a common framework and under a single generalist policy. These include a variety of gaits, traversing complex terrains (eg. stairs and slopes) as well as previously unseen stepping-stones and narrow beams, and interacting with objects (eg. pushing buttons, tracking trajectories). Our framework acquires new behaviors more efficiently, simply by combining a task-specific high-level contact planner and the pre-trained generalist policy.

Video

Method

GeCCo samples from a huge range of contact configurations during training, enabling it to generalize to a variety of tasks at test time. These include different gaits, manipulating behaviours with any of the feet while balancing, and traversing different terrains. The policy is conditioned on the desired contact points for each foot and is trained with Deep Reinforcement Learning:

Results

GeCCo is highly robust and can achieve a variety of highly dynamic behaviours in sim and in the real world. Cycle through the tabs below to see some of the tasks it can perform:

Various Gaits

Here we show some of the gaits that GeCCo can achieve in the real-world:

Trot

Pace

Bound

Pronk

Transitioning between gaits

Complex Terrains

GeCCo can also traverse difficult terrains like stairs and slopes:

Across a sloped ramp

Across a large staircase

Across a narrow beam

Across smaller steps

Down-slope

Up-slope

Down-stairs

Up-stairs

Sparse terrains (zero-shot)

GeCCo can even traverse sparse terrains in zero-shot, without ever encountering them during training:

Double beam trot

Double beam pronk

Narrow beam trot

Narrow beam pronk

Stepping Stones (zero-shot)

Locomanipulation

In exactly the same way, we can specify 3D coordinates for any of the feet - enabling locomanipulation capabilities. We can further combine and seamlessly transition between locomotion and manipulation:

Trace a circle

Reach

Box push

Remove obstacle and walk

Button press

Barrel roll

Barrel push + trot

Barrel push + bound

BibTeX

@misc{atanassov2025geccogeneralistcontactconditioned,
      title={GeCCo -- a Generalist Contact-Conditioned Policy for Loco-Manipulation Skills on Legged Robots}, 
      author={Vassil Atanassov and Wanming Yu and Siddhant Gangapurwala and James Wilson and Ioannis Havoutis},
      year={2025},
      eprint={2509.17582},
      archivePrefix={arXiv},
      primaryClass={cs.RO},
      url={https://arxiv.org/abs/2509.17582}, 
}