Shafeef Omar

2026

1. 

   DiPLO: Self-distilled Diffusion Planning for Loco-manipulation

   Shafeef Omar, Sebastian Sanokowski, Elham Daneshmand, Dian Yu, Daniel Cremers, Cordelia Schmid, and Majid Khadiv

   arXiv preprint arXiv:2610.24847, 2026

   Abs arXiv Bib PDF Website Video

   DiPLO pairs a diffusion planner with an RL-trained whole-body tracker for humanoid loco-manipulation. Self-distillation removes the planner’s dependence on reference-motion conditioning, and reinforcement learning then improves its plans with the tracker held fixed. The resulting system plans from state history and object goals, generalizes to new goals, and executes pick-carry-place motions on a Unitree G1 without real-world fine-tuning.

   @article{omar2026diplo,
     author = {Omar, Shafeef and Sanokowski, Sebastian and Daneshmand, Elham and Yu, Dian and Cremers, Daniel and Schmid, Cordelia and Khadiv, Majid},
     title = {{DiPLO}: Self-distilled Diffusion Planning for Loco-manipulation},
     journal = {arXiv preprint arXiv:2610.24847},
     year = {2026},
   }

2. 

   MotionDisco: Motion Discovery for Extreme Humanoid Loco-Manipulation

   Ilyass Taouil*, Michal Ciebielski*, Shafeef Omar*, Haizhou Zhao, Angela Dai, Aaron M. Johnson, and Majid Khadiv

   arXiv preprint arXiv:2606.06139, 2026

   Equal contribution: Ilyass Taouil, Michal Ciebielski, and Shafeef Omar.

   Abs arXiv Bib PDF Website Video

   MotionDisco discovers humanoid loco-manipulation motions without demonstrations or teleoperation. Language-model-guided evolutionary search proposes contact interactions, while a trajectory optimizer checks and refines the resulting whole-body motions. RL tracking policies transfer the discovered skills to a real humanoid robot.

   @article{taouil26motiondisco,
     author = {Taouil*, Ilyass and Ciebielski*, Michal and Omar*, Shafeef and Zhao, Haizhou and Dai, Angela and Johnson, Aaron M. and Khadiv, Majid},
     title = {{MotionDisco}: Motion Discovery for Extreme Humanoid Loco-Manipulation},
     journal = {arXiv preprint arXiv:2606.06139},
     year = {2026},
     note = {Equal contribution: Ilyass Taouil, Michal Ciebielski, and Shafeef Omar.},
   }

3. 

   DynaRetarget: Dynamically-Feasible Retargeting using Sampling-Based Trajectory Optimization

   Victor Dhédin*, Ilyass Taouil*, Shafeef Omar*, Dian Yu, Kun Tao, Angela Dai, and Majid Khadiv

   IEEE Robotics and Automation Letters (RA-L), 2026

   Accepted for publication. Equal contribution: Victor Dhédin, Ilyass Taouil, and Shafeef Omar.

   Abs arXiv Bib PDF Code Website Video

   DynaRetarget converts human demonstrations into dynamically feasible humanoid motions. Sampling-based trajectory optimization refines kinematic retargeting results over progressively longer horizons, handling variations in object mass, size, and geometry. The resulting motions support training humanoid control policies and generating synthetic loco-manipulation data.

   @article{dhedin26dynaretarget,
     author = {Dhédin*, Victor and Taouil*, Ilyass and Omar*, Shafeef and Yu, Dian and Tao, Kun and Dai, Angela and Khadiv, Majid},
     title = {DynaRetarget: Dynamically-Feasible Retargeting using Sampling-Based Trajectory Optimization},
     journal = {IEEE Robotics and Automation Letters (RA-L)},
     year = {2026},
     note = {Accepted for publication. Equal contribution: Victor Dhédin, Ilyass Taouil, and Shafeef Omar.},
   }

4. 

   Learning to Act Through Contact: A Unified View of Multi-Task Robot Learning

   Shafeef Omar*, and Majid Khadiv

   In Proceedings of The 8th Annual Learning for Dynamics and Control Conference, 2026

   Abs arXiv Bib HTML PDF Video

   This work represents locomotion and manipulation tasks as sequences of contact goals: where and when contact should occur, and which end-effectors should be active. A goal-conditioned RL policy learns multiple skills within each robotic setting, including quadrupedal gaits, humanoid locomotion, and bimanual object manipulation. Explicit contact goals improve generalization to unseen tasks.

   @inproceedings{omar25contact,
     author = {Omar*, Shafeef and Khadiv, Majid},
     title = {Learning to Act Through Contact: A Unified View of Multi-Task Robot Learning},
     booktitle = {Proceedings of The 8th Annual Learning for Dynamics and Control Conference},
     series = {Proceedings of Machine Learning Research},
     volume = {331},
     pages = {1199--1211},
     publisher = {PMLR},
     year = {2026},
   }

2023

1. 

   SafeSteps: Learning Safer Footstep Planning Policies for Legged Robots via Model-Based Priors

   Shafeef Omar*, Lorenzo Amatucci, Victor Barasuol, Giulio Turrisi, and Claudio Semini

   In IEEE-RAS International Conference on Humanoid Robots, 2023

   Abs arXiv Bib PDF Website

   We present a footstep planning policy for quadrupedal locomotion that is able to directly take into consideration a-priori safety information in its decisions. At its core, a learning process analyzes terrain patches, classifying each landing location by its kinematic feasibility, shin collision, and terrain roughness. This information is then encoded into a small vector representation and passed as an additional state to the footstep planning policy, which furthermore proposes only safe footstep location by applying a masked variant of the Proximal Policy Optimization (PPO) algorithm [1]. The performance of the proposed approach is shown by comparative simulations on an electric quadruped robot walking in different rough terrain scenarios. We show that violations of the above safety conditions are greatly reduced both during training and the successive deployment of the policy, resulting in an inherently safer footstep planner. Furthermore, we show how, as a byproduct, fewer reward terms are needed to shape the behavior of the policy, which in return is able to achieve both better final performances and sample efficiency.

   @inproceedings{omar23humanoids,
     author = {Omar*, Shafeef and Amatucci, Lorenzo and Barasuol, Victor and Turrisi, Giulio and Semini, Claudio},
     title = {SafeSteps: Learning Safer Footstep Planning Policies for Legged Robots via Model-Based Priors},
     booktitle = {IEEE-RAS International Conference on Humanoid Robots},
     year = {2023},
   }