![[2602.14363] AdaptManip: Learning Adaptive Whole-Body Object Lifting and Delivery with Online Recurrent State Estimation](https://arxiv.org/static/browse/0.3.4/images/arxiv-logo-fb.png){kind=logo}
[2602.14363] AdaptManip: Learning Adaptive Whole-Body Object Lifting and Delivery with Online Recurrent State Estimation
Summary
The paper presents AdaptManip, an autonomous framework for humanoid robots to perform object lifting and delivery using reinforcement learning, enhancing adaptability and performance over traditional methods.
Why It Matters
As robotics continues to evolve, the ability to perform complex tasks autonomously is crucial. AdaptManip's approach to integrating navigation and manipulation without human demonstrations represents a significant advancement in robotic capabilities, potentially impacting industries such as logistics and healthcare.
Key Takeaways
- AdaptManip uses reinforcement learning for robust object manipulation without human demonstrations.
- The framework includes a real-time object state estimator and a whole-body base policy for effective locomotion.
- Experimental results show significant improvements in adaptability and success rates compared to imitation learning methods.
- The system is capable of fully autonomous navigation and delivery in real-world scenarios.
- Accurate object state estimation enhances manipulation performance even in challenging conditions.
Computer Science > Robotics arXiv:2602.14363 (cs) [Submitted on 16 Feb 2026] Title:AdaptManip: Learning Adaptive Whole-Body Object Lifting and Delivery with Online Recurrent State Estimation Authors:Morgan Byrd, Donghoon Baek, Kartik Garg, Hyunyoung Jung, Daesol Cho, Maks Sorokin, Robert Wright, Sehoon Ha View a PDF of the paper titled AdaptManip: Learning Adaptive Whole-Body Object Lifting and Delivery with Online Recurrent State Estimation, by Morgan Byrd and 7 other authors View PDF Abstract:This paper presents Adaptive Whole-body Loco-Manipulation, AdaptManip, a fully autonomous framework for humanoid robots to perform integrated navigation, object lifting, and delivery. Unlike prior imitation learning-based approaches that rely on human demonstrations and are often brittle to disturbances, AdaptManip aims to train a robust loco-manipulation policy via reinforcement learning without human demonstrations or teleoperation data. The proposed framework consists of three coupled components: (1) a recurrent object state estimator that tracks the manipulated object in real time under limited field-of-view and occlusions; (2) a whole-body base policy for robust locomotion with residual manipulation control for stable object lifting and delivery; and (3) a LiDAR-based robot global position estimator that provides drift-robust localization. All components are trained in simulation using reinforcement learning and deployed on real hardware in a zero-shot manner. Experimental resu...
