Research artifacts linked to products and companies tracked by Korthos.
Apptronik describes Robot Park as a data-collection and training environment where Apollo 2 robots collect real-world task data for humanoid intelligence. The source links Apollo 2, teleoperation, simulation, and Google DeepMind Gemini Robotics model development.
In the field of robot learning, large-scale and diverse demonstration trajectories provide the fundamental basis for enhancing robotic manipulation ability. We introduce RoboTacDex, a large, multi-modal, and diverse dataset of dexterous manipulation behaviors performed with a humanoid robot.
Flexion reports a long-horizon humanoid autonomy stack that composes mission-level reasoning, perception, manipulation, locomotion, recovery, and runtime infrastructure. In a 16-step mission evaluation, the post says supervised fine-tuning reached 38% end-to-end completion while SFT plus RL reached 90%.
Perception-based humanoid loco-manipulation requires connecting egocentric observations and task instructions to whole-body motion. Learning this mapping requires synchronized egocentric images, language commands, and robot-compatible kinematic trajectories, yet no existing data source provides this complete tuple at scale.
While current Behavior Foundation Models (BFMs) provide robust control priors for humanoids, they only execute pre-defined reference motions. As a result, they are vulnerable to environmental shifts and incapable of reactive whole-body coordination.
The XDOF-linked paper introduces Warp-Augmented Relative Progress, using time-warped demonstration clips to learn dense progress rewards without human subtask labels. On widened, noisier folding data, WARP-BC maintained 19/20 successful folds where vanilla behavior cloning dropped to 2/20, and it improved bottle-placement throughput on a second physical task.
Model Predictive Control (MPC) is the standard predictive layer in hierarchical quadruped controllers, but the per-cycle QP solve limits the update rate achievable on embedded processors. Because legged gaits revisit a bounded region of state space, MPC solutions admit caching and reuse.
Quadruped robots have achieved remarkable locomotion, yet their behavioral repertoire remains confined to a few gaits--far from the expressive, companion-like presence long envisioned for them. Attempts to import the humanoid recipe of large-scale motion data have inherited one tacit assumption: that robot motion must first pass through an animal body, making data collection dependent on cooperative animals, reconstruction fragile across species, and retargeting ill-posed across incompatible morphologies.
This paper presents a hierarchical control framework using model predictive control (MPC) and reinforcement learning (RL) for active roll control to manage lateral load transfer during autonomous racing of a wheeled quadruped. The framework integrates offline time-optimal raceline generation, an online MPC planner that actively minimizes the lateral Load Transfer Ratio (LTR), and a low-level, whole-body RL policy deployed directly onto the robot's 16 actuators.
High-level humanoid planners often output sparse task-space, low-rate trajectories, whereas whole-body controllers run at high frequency.
Humanoid robots could take on physically demanding, hazardous, and repetitive work in spaces built for humans. However, a useful robot for these spaces must coordinate locomotion, whole body motion, perception, contact, and operator supervision.
StairMaster is a reinforcement-learning framework for quadruped locomotion on risky hollow stairs, using depth perception, recurrent memory, and sim-to-real sensor modeling.
RoboAtlas is a contextual Active SLAM framework that combines frontier exploration, 3D semantic mapping, global semantic reasoning, and egocentric VLM reasoning.
TaskNPoint is a training protocol for teaching humanoid robots dynamic skills from short human video demonstrations and task-critical interaction windows.
Humanoid robots require whole-body controllers that are both robust and precise in contact-rich environments. While deep reinforcement learning (RL) achieves robust stability, its behavior is tightly coupled to the training objective and command interface, making it difficult to add new feedback objectives without retraining.
Universal quadrupedal locomotion remains limited by the difficulty of integrating perception across diverse robot morphologies. State-of-the-art controllers rely on single-robot training or blind policies that omit real-time perception, leading to poor cross-embodiment generalization.
Augmented Reality (AR) can improve collocated human-robot collaboration by making robot state and intent visible and enabling intuitive control, yet large, visually diverse environments like the outdoors challenge both interaction and content legibility, especially at long distances and beyond visual line of sight.
The paper proposes a real-time shared-control layer that reshapes teleoperated arm motion toward human-like end-effector trajectories while preserving operator intent.
Operating in complex real-world environments requires robots to understand their surroundings on a functional semantic level.
This paper presents a modular training-free framework for zero-shot, language-guided robotic manipulation in semi-structured environments. The architecture bridges the gap between high-level reasoning and low-level kinematics by decomposing the vision-action pipeline into three stages: visual perception, semantic interpretation, and task execution.
The paper reports experimental pose-accuracy testing of the UR20 robot from Universal Robots. The experiments follow ISO 9283 guidelines and use an OptiTrack vision system to record robot positions.
Humanoid local navigation in cluttered environments must jointly resolve obstacle avoidance, sparse-goal recovery, and stable whole-body locomotion under short-range and partially observable sensing.
Text-conditioned motion generation is a promising interface for programming humanoid robots, yet current generators are often trained on human motion datasets retargeted to robot morphologies. Although such data provides rich semantic and kinematic priors, it fails to capture the nuances of whole-body tracking controllers, including balance, contact dynamics, actuation limits, and controller-specific failure modes.
The arXiv paper coordinates body and hand priors for continuous dexterous loco-manipulation on Unitree G1.
Multi-task robot manipulation policies are challenging to learn from demonstration because traditionally a single network must select among qualitatively different action modes from a multimodal demonstration distribution, conditioned on language and visual context. A wrong mode selection means executing the wrong task or an action infeasible in the scene.
The paper builds a deep-reinforcement-learning locomotion controller and autonomous navigation system for the commercially available Go2-W.
The paper studies dual-robot understanding via efficient teaching with Unitree G1. The artifact names Unitree G1 in its metadata, making the humanoid product binding explicit.
Studies whether filtered egocentric human videos with pseudo-action labels can outperform matched real-robot teleoperation data for embodied pretraining. Reports better action-prediction validation loss and higher real-robot task success across in-distribution and out-of-distribution settings. Supports a scalable pretrain-on-human-video, adapt-with-robot-data paradigm for embodied robot policies. Studies whether filtered egocentric human videos with pseudo-action labels can outperform matched real-robot teleoperation data for embodied pretraining. Reports better action-prediction validation loss and higher real-robot task success across in-distribution and out-of-distribution settings. Supports a scalable pretrain-on-human-video, adapt-with-robot-data paradigm for embodied robot policies. Timely embodied-data result that directly informs robot-data scaling strategy and human-video pretraining.
Legged robots are increasingly deployed in forests for ecological surveying and monitoring, yet their autonomy is often interrupted consequent to the challenges posed in traversing forest environments.
Autonomous navigation of quadrupedal robots in diverse environments fundamentally relies on resilient Simultaneous Localization and Mapping (SLAM). While visual-inertial SLAM has matured across wheeled, handheld, and aerial platforms, a critical evaluation gap remains regarding how hardware-level sensor configurations affect performance under the aggressive dynamics of legged locomotion.
Human demonstrations, which can be collected at scale and naturally capture active hand-eye coordination, are a promising data source for learning humanoid loco-manipulation.
Introduces a closed-loop physical autoresearch system that lets coding agents reset tasks, run real-robot rollouts, verify outcomes, and refine policies. Reports 99% pass@8 success on dexterous manipulation tasks including pin insertion, zip-tie cutting, and GPU insertion. Shows fleet scaling across one, four, and eight agent-robot teams for faster real-world policy improvement. Introduces a closed-loop physical autoresearch system that lets coding agents reset tasks, run real-robot rollouts, verify outcomes, and refine policies. Reports 99% pass@8 success on dexterous manipulation tasks including pin insertion, zip-tie cutting, and GPU insertion. Shows fleet scaling across one, four, and eight agent-robot teams for faster real-world policy improvement. Concrete real-robot system for agent-driven robot policy improvement, with project page, paper link, and hardware-fleet evaluation.
Introduces Qwen-RobotNav, Qwen-RobotManip, and Qwen-RobotWorld as a foundation-model suite for physical-world intelligence. Covers language-conditioned navigation, cross-embodiment manipulation, and video world modeling for robot scenarios. Provides technical reports and model resources from the Qwen research program. Introduces Qwen-RobotNav, Qwen-RobotManip, and Qwen-RobotWorld as a foundation-model suite for physical-world intelligence. Covers language-conditioned navigation, cross-embodiment manipulation, and video world modeling for robot scenarios. Provides technical reports and model resources from the Qwen research program. Major foundation-model release from Alibaba covering robot navigation, manipulation, and world modeling.
Diffusion and flow-based generative policies provide a powerful policy class for reinforcement learning by inducing rich stochastic exploration through iterative action generation.
APEX is a plug-and-play execution layer that reconstructs dynamically feasible references from learned policy outputs and adapts online to reduce controller tracking error.
This paper covers SemGeoNav paper evaluates safety-guided visual navigation on Unitree Go2 for Go2 in Navigation.
Humanoids deployed in human-centered environments must handle force-interactive tasks, where external contacts introduce unexpected disturbances that disrupt locomotion accuracy and stability. Existing learning-based approaches rely on broad domain randomization, task-specific force objectives, or learning-based force estimators from motion history, each of which compromises accuracy, task transferability, or out-of-distribution (OOD) robustness.
Humanoid robots performing in-field manipulation tasks, such as robotic apple harvesting, face severe energy constraints that directly limit the number of reaching motions that can be executed per battery charge. This paper presents an end-to-end, energy-aware reinforcement learning framework for the 7-degree-of-freedom left arm of the Unitree~G1 humanoid robot, combining a physics-based, experimentally identified electrical power model with a Soft Actor-Critic (SAC) policy trained in a Pinocchio-based rigid-body d
Learning-based quadrupedal locomotion typically relies on complex reward formulations that entangle task specification, operational limits, gait preference, and terrain adaptation within a single optimization objective.
Deep reinforcement learning has shown strong potential for robot navigation, but its practical deployment is still limited by the long wall-clock cost of policy training.
Accurate prediction of electrical power consumption is essential for energy-aware motion planning, battery management, and thermal monitoring in battery-powered humanoid robots. This letter presents a physics-based, linear-in-parameters model for the electrical power consumption of the seven-degree-of-freedom left arm of the Unitree~G1 humanoid robot.
Open-sources a LiDAR localization stack adapted for humanoid platforms such as Unitree G1-style robots. Builds on FAST-LIO with humanoid-oriented deployment changes and ROS/C++ prerequisites. Adds practical perception/navigation infrastructure for stable humanoid operation in unstructured environments. Open-sources a LiDAR localization stack adapted for humanoid platforms such as Unitree G1-style robots. Builds on FAST-LIO with humanoid-oriented deployment changes and ROS/C++ prerequisites. Adds practical perception/navigation infrastructure for stable humanoid operation in unstructured environments. Concrete GitHub release for humanoid perception and navigation support infrastructure.
Floating-base robots must balance under rigid contact constraints while interacting safely with humans. Existing whole-body control~(WBC) frameworks allocate the full joint space to locomotion or rely on fixed-gain impedance feedback that accumulates steady-state error under sustained physical human--robot interaction~(pHRI) forces.
The recent popularity of robotics, combined with the steadily decreasing cost of robotic hardware, has lowered the entry barrier to robotics research and enabled rapid advancements in the field. One of the primary examples is the Unitree Go2 quadruped robot, which is often used by researchers in the areas of locomotion, navigation, control, and others.
The paper presents an impedance MPC framework for dexterous fingers and evaluates scaling to a 16-DOF LEAP Hand simulation.
Open-sources a hardware/software framework for scalable robot learning with human-operated data collection and inspection pipelines. Pairs the framework with a Hugging Face G0 dataset surface for multimodal dexterous manipulation data. Targets reduced real-robot data requirements through interfaces, quality control, and better data ratios. Open-sources a hardware/software framework for scalable robot learning with human-operated data collection and inspection pipelines. Pairs the framework with a Hugging Face G0 dataset surface for multimodal dexterous manipulation data. Targets reduced real-robot data requirements through interfaces, quality control, and better data ratios. Major open-source dataset/framework release aimed directly at robot-learning data bottlenecks.
Frames dexterous articulated-tool manipulation as an animation-style pipeline for tools such as tongs, pliers, clothespins, and syringes. Combines procedural grasp keyframes, motion planning, and reinforcement learning for trajectories. Project page reports zero-shot sim-to-real transfer with minimal per-tool user input and foundation-model point-cloud observations. Frames dexterous articulated-tool manipulation as an animation-style pipeline for tools such as tongs, pliers, clothespins, and syringes. Combines procedural grasp keyframes, motion planning, and reinforcement learning for trajectories. Project page reports zero-shot sim-to-real transfer with minimal per-tool user input and foundation-model point-cloud observations. Fresh dexterous manipulation project with strong sim-to-real framing and project-page evidence.
Depth estimation has numerous medical and surgical applications. We benchmark four depth sensors on a porcine bone specimen, a porcine belly specimen, and a silicone kidney phantom using stylus-sampled references.
A structured directory for robotics research artifacts linked to companies, products, capabilities, and source events: datasets, models, benchmarks, repositories, simulators, and technical releases enriched from Korthos monitoring.
This is not an exhaustive database of robotics papers. It surfaces research artifacts linked to products and companies tracked by Korthos.