arXiv:2511.04679v1 Announce Type: cross 
Abstract: Humanoid robots are expected to operate in human-centered environments where safe and natural physical interaction is essential. However, most recent reinforcement learning (RL) policies emphasize rigid tracking and suppress external forces. Existing impedance-augmented approaches are typically restricted to base or end-effector control and focus on resisting extreme forces rather than enabling compliance. We introduce GentleHumanoid, a framework that integrates impedance control into a whole-body motion tracking policy to achieve upper-body compliance. At its core is a unified spring-based formulation that models both resistive contacts (restoring forces when pressing against surfaces) and guiding contacts (pushes or pulls sampled from human motion data). This formulation ensures kinematically consistent forces across the shoulder, elbow, and wrist, while exposing the policy to diverse interaction scenarios. Safety is further supported through task-adjustable force thresholds. We evaluate our approach in both simulation and on the Unitree G1 humanoid across tasks requiring different levels of compliance, including gentle hugging, sit-to-stand assistance, and safe object manipulation. Compared to baselines, our policy consistently reduces peak contact forces while maintaining task success, resulting in smoother and more natural interactions. These results highlight a step toward humanoid robots that can safely and effectively collaborate with humans and handle objects in real-world environments.

تقديم GentleHumanoid يمثل تقدمًا كبيرًا في مجال الروبوتات البشرية، حيث يركز على تحسين التفاعلات الآمنة والطبيعية بين الروبوتات والبشر أو الأشياء. على عكس طرق التعلم المعزز التقليدية التي تعطي الأولوية للتحكم الصارم، يبرز GentleHumanoid الامتثال، مما يسمح للروبوتات بالتكيف مع قوى متغيرة أثناء التفاعلات. هذا التطور مهم لأنه يمهد الطريق لتعاون أكثر فعالية وبديهية بين البشر والروبوتات في البيئات اليومية، مما يجعل الروبوتات أكثر تنوعًا وسهولة في الاستخدام.

La introducción de GentleHumanoid marca un avance significativo en el campo de la robótica humanoide, centrándose en mejorar las interacciones seguras y naturales entre robots y humanos u objetos. A diferencia de los métodos tradicionales de aprendizaje por refuerzo que priorizan el control rígido, GentleHumanoid enfatiza la conformidad, permitiendo que los robots se adapten a fuerzas variables durante las interacciones. Este desarrollo es crucial, ya que allana el camino para una colaboración más efectiva e intuitiva entre humanos y robots en entornos cotidianos, haciendo que los robots sean más versátiles y fáciles de usar.

L'introduction de GentleHumanoid représente une avancée significative dans le domaine de la robotique humanoïde, en mettant l'accent sur l'amélioration des interactions sûres et naturelles entre les robots et les humains ou les objets. Contrairement aux méthodes d'apprentissage par renforcement traditionnelles qui privilégient le contrôle rigide, GentleHumanoid souligne la conformité, permettant aux robots de s'adapter à des forces variées lors des interactions. Ce développement est crucial car il ouvre la voie à une collaboration plus efficace et intuitive entre les humains et les robots dans les environnements quotidiens, rendant les robots plus polyvalents et conviviaux.

The introduction of GentleHumanoid marks a significant advancement in the field of humanoid robotics, focusing on enhancing safe and natural interactions between robots and humans or objects. Unlike traditional reinforcement learning methods that prioritize rigid control, GentleHumanoid emphasizes compliance, allowing robots to adapt to varying forces during interactions. This development is crucial as it paves the way for more effective and intuitive human-robot collaboration in everyday environments, making robots more versatile and user-friendly.

GentleHumanoid: Learning Upper-body Compliance for Contact-rich Human and Object Interaction

<p>General-purpose robots remain rare not for a lack of hardware but because we still can&rsquo;t give machines the physical intuition humans learn through experience</p>

لا تزال الروبوتات الشبيهة بالبشر والذكاء الاصطناعي المجسد تواجه تحديات كبيرة في التطبيقات الواقعية، ويرجع ذلك أساسًا إلى عدم قدرتها على تكرار الحدس الجسدي الذي يطوره البشر من خلال التجربة. وقد أدت هذه القيود إلى تقدم بطيء في نشر هذه التقنيات على الرغم من التقدم في الأجهزة.

Los robots humanoides de propósito general y la IA encarnada siguen enfrentando desafíos significativos en aplicaciones del mundo real, principalmente debido a su incapacidad para replicar la intuición física que los humanos desarrollan a través de la experiencia. Esta limitación ha resultado en un progreso lento en el despliegue de estas tecnologías a pesar de los avances en hardware.

Les robots humanoïdes polyvalents et l'IA incarnée continuent de faire face à des défis significatifs dans les applications réelles, principalement en raison de leur incapacité à reproduire l'intuition physique que les humains développent par l'expérience. Cette limitation a entraîné une progression lente dans le déploiement de ces technologies malgré les avancées matérielles.

General-purpose humanoid robots and embodied AI continue to face significant challenges in real-world applications, primarily due to their inability to replicate the physical intuition that humans develop through experience. This limitation has resulted in a slow progression in the deployment of these technologies despite advancements in hardware.

Why Humanoid Robots and Embodied AI Still Struggle in the Real World

arXiv:2508.05876v2 Announce Type: replace 
Abstract: We develop a Markov decision process (MDP) framework to autonomously make guidance decisions for satellite collision avoidance maneuver (CAM) and a reinforcement learning policy gradient (RL-PG) algorithm to enable direct optimization of guidance policy using historic CAM data. In addition to maintaining acceptable collision risks, this approach seeks to minimize the average propellant consumption of CAMs by making early maneuver decisions. We model CAM as a continuous state, discrete action and finite horizon MDP, where the critical decision is determining when to initiate the maneuver. The MDP models decision rewards using analytical models of collision risk, propellant consumption, and transit orbit geometry. By deciding to maneuver earlier than conventional methods, the Markov policy effectively favors CAMs that achieve comparable rates of collision risk reduction while consuming less propellant. Using historical data of tracked conjunction events, we verify this framework and conduct an extensive parameter-sensitivity study. When evaluated on synthetic conjunction events, the trained policy consumes significantly less propellant overall and per maneuver in comparison to a conventional cut-off policy that initiates maneuvers 24 hours before the time of closest approach (TCA). On historical conjunction events, the trained policy consumes more propellant overall but consumes less propellant per maneuver. For both historical and synthetic conjunction events, the trained policy is slightly more conservative in identifying conjunctions events that warrant CAMs in comparison to cutoff policies.

تم تطوير إطار جديد لعملية اتخاذ القرار ماركوف (MDP) لتوجيه المناورات الذاتية لتجنب تصادم الأقمار الصناعية (CAM). يستخدم هذا الإطار خوارزمية تدرج السياسة للتعلم المعزز لتحسين سياسات التوجيه بناءً على بيانات تاريخية لـ CAM، بهدف تقليل مخاطر التصادم مع تقليل استهلاك الوقود من خلال اتخاذ قرارات المناورة في وقت مبكر.

Se ha desarrollado un nuevo marco de proceso de decisión de Markov (MDP) para guiar de manera autónoma las maniobras de evitación de colisiones de satélites (CAM). Este marco utiliza un algoritmo de gradiente de política de aprendizaje por refuerzo para optimizar las políticas de guía basadas en datos históricos de CAM, con el objetivo de reducir los riesgos de colisión mientras minimiza el consumo de propulsor mediante decisiones de maniobra más tempranas.

Un nouveau cadre de processus décisionnel de Markov (MDP) a été développé pour guider de manière autonome les manœuvres d'évitement de collision des satellites (CAM). Ce cadre utilise un algorithme de gradient de politique d'apprentissage par renforcement pour optimiser les politiques de guidage basées sur des données historiques de CAM, visant à réduire les risques de collision tout en minimisant la consommation de propulseur grâce à des décisions de manœuvre plus précoces.

A new Markov decision process (MDP) framework has been developed to autonomously guide satellite collision avoidance maneuvers (CAMs). This framework utilizes a reinforcement learning policy gradient algorithm to optimize guidance policies based on historical CAM data, aiming to reduce collision risks while minimizing propellant consumption through earlier maneuver decisions.

A Markov Decision Process Framework for Early Maneuver Decisions in Satellite Collision Avoidance

One More Thing in AI – Your Shortcut to AI Mastery

GentleHumanoid: Learning Upper-body Compliance for Contact-rich Human and Object Interaction

Was this article worth reading? Share it

One More Thing in AI

Humanize AI

GPTHumanizer

AIvilization

Grubby.AI

AI Humanizer

Ready to build your own newsroom?