arXiv:2606. 32027v1 Announce Type: cross Abstract: Reward design remains a central bottleneck for autonomous robot policy improvement, especially in long-horizon manipulation tasks where sparse success labels provide too little signal and binary preferences collapse many competing notions of quality into one ambiguous signal.
By Marcel Torne, Anubha Mahajan, Abhijnya Bhat, Chelsea Finn
arXiv:2608. 02951v1 Announce Type: cross Abstract: Preference-based reinforcement learning (PbRL) for general stochastic MDPs often requires training a reward model.
By Evan Assmus, Qining Zhang, Lei Ying
The paper introduces Residual Reward Models (RRM) to enhance preference‑based reinforcement learning (PbRL) in robotics. RRMs decompose the true reward into a prior component—such as a heuristic, language‑generated, or IRL‑derived reward—and a learned residual that is trained with human preferences. Experiments on Meta‑World, DM‑Control, and a physical Franka Panda robot show that RRMs markedly improve sample efficiency and accelerate policy learning compared to standard PbRL methods.
By Chenyang Cao, Miguel Rogel-Garc\'ia, Mohamed Nabail, Xueqian Wang, Nicholas Rhinehart
The paper introduces Reward Ensemble under Confidence (REC), a probabilistic reward learning framework for preference-based reinforcement learning that models per‑timestep reward uncertainty using an ensemble of distributional reward models. REC incorporates uncertainty into the preference loss and uses model disagreement to drive exploration, achieving 88.4% of shaped‑reward performance on acrobatic quadrotor control versus 55.2% with standard Preference PPO. The authors train policies in simulation and transfer them zero‑shot to real quadrotors, demonstrating complex acrobatic maneuvers learned solely from human preference feedback, and validate REC on a continuous‑control benchmark.
By Colin Merk, Ismail Geles, Jiaxu Xing, Angel Romero, Giorgia Ramponi, Davide Scaramuzza
PrefPI (Preference-Guided Policy Iteration) is an iterative framework that steers pretrained generative robot policies using only relative preferences over self-generated trajectories. It treats preference learning as preference-conditioned generative modeling, where preferred trajectories define a conditional distribution whose density ratio with the broader behavior prior yields an implicit preference signal amplified by classifier-free guidance (CFG). By repeatedly applying this preference-conditioned modeling and guidance, PrefPI iteratively improves policies, enabling access to behaviors that were rarely or never observed under the initial policy, and achieves significant behavioral shifts such as increasing object transport height from 10.7 cm to 19.8 cm on real hardware with only 150 preference-labeled trajectories.
By Seungeun Rho, Wontaek Kim, Danfei Xu, Sehoon Ha
arXiv:2610.01260v1 Announce Type: cross
Abstract: Quadrupedal locomotion requires balancing conflicting objectives such as command tracking, stability, and energy efficiency, yet conventional reinfor...
By Amr Mousa, Rifny Rachman, Neil Karavis, Michele Caprio, Richard Allmendinger