Al Jaber Mahmud

Al Jaber Mahmud, Weizi Li, Xuan Wang

Mutual adaptation can enhance overall task performance in human-robot co-transportation by integrating both the robot's and the human's understanding of the environment. While human modeling helps capture humans' subjective preferences, two challenges persist: (i) the uncertainty of human preference parameters and (ii) the need to balance adaptation strategies that benefit both humans and robots. In this paper, we propose a unified framework to address these challenges and improve task performance through mutual adaptation. First, instead of relying on fixed parameters, we model a probability distribution of human choices by incorporating a range of uncertain human preference parameters. Building on this, we introduce a time-varying stubbornness measure and a coordinated planning model, which allows either the robot to lead the team's trajectory or, if a human's preferred path conflicts with the robot's plan and their stubbornness exceeds a threshold, the robot to transition to following the human. Finally, we introduce a pose optimization strategy for low-level control to mitigate the uncertain human behaviors when they are leading. To validate the framework, we design and perform a study with human feedback from twenty human participants. We then demonstrate, through simulations, the effectiveness of our models in enhancing task performance with mutual adaptation and pose optimization.

Al Jaber Mahmud, Xuan Wang

We study rigid-body motion planning through multiple sequential narrow openings, which requires long-horizon geometric reasoning because the configuration used to traverse an early opening constrains the set of reachable configurations for subsequent ones. To achieve this, we propose a geometry-aligned large language model (LLM) fine-tuning framework that generates fixed-length, machine-readable waypoint sequences that are both geometrically feasible and coordinated across openings. Our approach uses a bi-level training pipeline. First, we perform failure-driven LoRA supervised fine-tuning (SFT) on human demonstrations, which incorporates structured failure feedback to teach the model common failure modes and enforce the output format. Second, we refine the same LoRA adapters using Group Relative Policy Optimization (GRPO) with geometric verification: each sampled waypoint sequence is densified by a model-based planner and scored with a deterministic geometry-derived reward to achieve continuous-motion feasibility. To validate the effectiveness of our proposed method, we provide both quantitative and qualitative results from simulations. Our method achieves the highest success rate in both in-distribution and out-of-distribution environments and qualitatively exhibits long-horizon geometric reasoning by selecting exit poses that facilitate entry into subsequent openings.