Jekyll2026-09-09T08:36:02+00:00https://abraranwar.github.io/feed.xmlAbrar AnwarAbrar Anwar is a robotics researcher at NVIDIA on the Isaac Loco-Manipulation team, working on reinforcement learning post-training of robot foundation models. Previously a PhD student at USC with Jesse Thomason on language-guided robot learning and evaluation.Anticipating Unintended Robot Behaviors with Side Effect Critics2026-09-09T00:00:00+00:002026-09-09T00:00:00+00:00https://abraranwar.github.io/side-effectsWe introduce the concept of Side Effect Critics to anticipate and mitigate unintended robot behaviors when deploying learned policies in real-world environments.]]>Mechanistic Finetuning of Vision-Language-Action Models via Few-Shot Demonstrations2026-09-01T00:00:00+00:002026-09-01T00:00:00+00:00https://abraranwar.github.io/steeringRobotic Steering is a mechanistic, few-shot finetuning approach for vision-language-action models that selectively adapts task-specific attention heads, achieving more robust, efficient, and interpretable robot learning than LoRA across diverse tasks.]]>SAFECAST: Robust Failure Detection for VLA Policies with Contrast-Set Training and Calibration2026-08-04T00:00:00+00:002026-08-04T00:00:00+00:00https://abraranwar.github.io/safecastVLA policies fail under deployment-time shifts like clutter, lighting changes, and reworded instructions. SAFECAST uses contrast set perturbations to train and calibrate hidden-state risk probes, improving failure detection ROC-AUC over state-of-the-art baselines on both real-world DROID and LIBERO simulation.]]>DiPS: Dialogue Policy Selection for High-Stakes Persuasion Agents2026-07-02T00:00:00+00:002026-07-02T00:00:00+00:00https://abraranwar.github.io/dipsDiPS is a Q-learning framework that dynamically selects persuasion strategies based on the evolving conversational context. In a fire-rescue evacuation scenario, DiPS achieves higher persuasion success than zero-shot LLM and RAG-augmented approaches in both simulated and real human interactions.]]>Robometer: Scaling General-Purpose Robotic Reward Models via Trajectory Comparisons2026-06-01T00:00:00+00:002026-06-01T00:00:00+00:00https://abraranwar.github.io/robometerRobometer is a general-purpose, video-language-input dense reward model trained on RBM-1M, a dataset of over 1M trajectories spanning 21 robot embodiments. It improves robot learning across online RL, offline RL, model-based RL, failure detection, and data retrieval for imitation learning.]]>Isaac Lab: A GPU-Accelerated Simulation Framework for Multi-Modal Robot Learning2025-11-06T00:00:00+00:002025-11-06T00:00:00+00:00https://abraranwar.github.io/isaaclabI helped introduce approaches for VLA post-training with IsaacLab!]]>RobotFleet: An Open-Source Framework for Centralized Multi-Robot Task Planning2025-10-12T00:00:00+00:002025-10-12T00:00:00+00:00https://abraranwar.github.io/robotfleetRobotFleet is an open-source, extensible framework that introduces a centralized, modular autonomy stack to simplify scalable planning, scheduling, and execution for heterogeneous multi-robot fleets in open-world tasks.]]>Efficient Evaluation of Multi-Task Robot Policies With Active Experiment Selection2025-08-28T00:00:00+00:002025-08-28T00:00:00+00:00https://abraranwar.github.io/seq-evalThe space of language commands a robot can execute grows combinatorially with scene complexity. Evaluating a robot on this large domain is impractical + takes time, so we introduce contrast sets for robots to make small perturbations to test instances. This leads to good test set estimation and less experimenter effort.]]>ReWiND: Language-Guided Rewards Teach Robot Policies without New Demonstrations2025-08-28T00:00:00+00:002025-08-28T00:00:00+00:00https://abraranwar.github.io/rewindWe design ReWiND rewards that use language-guided rewards to train bimanual arms on OOD tasks in 1 hour!
We use offline-to-online, lang-conditioned, visual RL on action-chunked transformers on a real robot and in simulation!]]>ReMEmbR: Building and Reasoning Over Long-Horizon Spatio-Temporal Memory for Robot Navigation2025-05-20T00:00:00+00:002025-05-20T00:00:00+00:00https://abraranwar.github.io/remembrTo tackle long-horizon spatio-temporal memory, we introduce a retrieval-based approach for building and reasoning over memory.
We show improved performance on planning and embodied question answering given a long video history.]]>