Researchers at the Grainger College of Engineering at the University of Illinois Urbana-Champaign combined reinforcement learning with control theory to propose a contraction-aware reinforcement learning method, CARL, which guarantees robot stability while improving tracking performance by up to 15% compared with existing methods.
Ken Goldberg shared the work of his postdoc Shuangyu0_0Xie: she found that Gemini 4 Argon performed well on spatial robotics queries, so she used Claude to reverse-engineer its skill library and implemented the robotics harness described in the Gemini 4 Argon report. The harness enabled Gemini to complete a zero-shot bimanual handover, and it is now open source.
SuperMimic trains a humanoid full-body control policy on 129,514 motions, surpassing the previous best result of 128 GPUs over 7 days with 1 GPU in 3 hours, reducing GPU hours by over 1,000x. Pieter Abbeel shared it and called it a game changer, noting that the key lies in using a world model faster than conventional simulation and in collecting far more signal than policy gradients through backpropagation through time. The project page is https://supermimic.github.io, and the demo is at https://supermimic.github.io/demo/.
Just want to confirm: the SOTA for VLA is an 80% success rate, right?
ConfigWM proposes a configurable perception mechanism that uses task language descriptions as configuration signals to guide the mapping of visual embeddings into task-specific state spaces, selectively retaining task-relevant information and suppressing distractions. In environments with progressively increasing distraction, existing methods show a marked decline in planning success rate, while ConfigWM remains relatively stable, averaging 22.3% higher than the strongest competing method. Physical-quantity probing (Pearson r +0.365) and distracted-to-clean retrieval (+75.5%) also show consistent improvements.
Unitree Technology launched its new H2-X Kung Fu Legend (Guardian) humanoid robot on October 21. The only core specification officially confirmed so far is a maximum battery life of about 8 hours; full hardware specifications, pricing, and mass-production delivery timing have not yet been announced.
Differential Review traveled to Chongqing to speak with engineers from Shengtong Energy and Qiten Robotics, and introduced the Metaverse Intelligent Inspection solution for high-risk industrial inspection. The solution first uses reconstruction technologies such as 3D Gaussian to build a digital factory on a computer, and a large model automatically identifies instruments.
Li Long, a PhD student at Griffith University, Zhao Qichao, a PhD student at Tsinghua University's Institute for Interdisciplinary Information Sciences, and teams from Fudan University and Tongji University propose Spotter: an embodied model executes continuously and autonomously while a lightweight local model supervises in parallel, and the VLM intervenes to correct errors only when they occur, avoiding the robot waiting for VLM inference at every step.
In his GTC Berlin keynote, Jensen Huang announced that Nvidia and Agile Robots are launching the Agile ONE SN reference humanoid robot, standing 1.4 meters tall, weighing 40 kilograms, with 31 degrees of freedom, preloaded with the Isaac GR00T model and powered by Jetson Thor. Expected to launch in early 2027, it is intended to cut the time needed to set up robot research hardware and software from weeks to days.
Slot-JEPA decomposes JEPA's dense patch representations into stable semantic slots and time-varying spatial masks, providing a reusable object-structured interface. Built on this interface, the Transition Memory Predictor (TMP) stores visual transitions as compact masks, retrieves context using slot statistics, and uses unbalanced optimal transport (UOT) to supervise mask prediction and evaluate candidate plans.
The 2026 Physical AI and World Models Conference opened in Nanjing, where Agnes AI showcased its self-developed agent runtime framework, Agnes Harness, in the exhibition area. Through the device access protocol MHS, it extends full-modal model capabilities to hardware including the Go2 robot dog and the Car01 robot vehicle.
The study proposes OverSeer, a multi-agent world model that adds auxiliary camera views fixed in space alongside agents' egocentric views, serving as a proxy for a world-centric representation to improve temporal and cross-agent consistency. The auxiliary views are redundant and nearly static, so they can be efficiently compressed with a delta-VAE that encodes only changes relative to a static reference image, reducing the number of tokens required.
NVIDIA released and open-sourced the data engine Flora, which converts raw robot video into training data such as timestamped action descriptions, process rewards, and outcome rewards; it is already used to train the GR00T series models. Flora consists of an action description model, a temporal segmentation model, and a reward model, fine-tuned from the Qwen3.5 series of vision-language models, and can label 200,000 hours of data per day on 512 GB300 GPUs.
Why it matters: NVIDIA's open-source data engine cuts robot video labeling costs to about 1/125 of a general-purpose large model and provides several benchmark comparisons, which can be used to assess the substitution space in the data labeling segment.
NVIDIA and the UW Medicine Institute for Protein Design launched the open model RFD4-Proteina, combining programmable protein design with search capabilities. In four benchmark tests reported in a research draft—protein binding, small-molecule binding, enzyme scaffolds, and DNA binding—computational design success rates and generation throughput were both higher than the evaluated baselines.
At GTC Berlin, NVIDIA announced an expansion of the Isaac open-source robotics development platform, adding the Isaac OS runtime, the officially released Isaac Lab 3.0 learning framework, and an open-source reference workflow based on DGX Station.
Why it matters: The original text lists the runtime, learning framework, and open-source workflow newly added to the Isaac platform, allowing readers to judge the specific changes in the data-to-deployment pipeline.
A robotics project uses evolutionary search and AI-generated CAD in simulation to let a robot iteratively design and test tools such as grippers to improve task performance; the robot's motions still come from a predefined task controller. The next step is to validate on real hardware using an Arctos robot and a 3D printer, and the author is seeking suggestions for automatically removing gripper fingers from the print bed.
Three days after Geek+ invested 50 million yuan in Shenyi Robot for a 20% stake, Shenyi Robot, which was founded just nine months ago, unveiled its self-developed embodied foundation model SE-1 at the 2026 CIIF, along with Sculptor, a multi-task embodied operation robot focused on heavy-load logistics, and Edge, a dual-arm robot focused on fine industrial operations.
US embodied intelligence startup Eka Robotics announced the close of a $150 million Series A round at a post-money valuation of about $1.45 billion, co-led by a16z and F-Prime, with Samsung Next and others participating.
PhysisServe, a physical AI inference system jointly released by Inverse Matrix, vLLM, the Beijing Academy of Artificial Intelligence, and Peking University, completes one chunk every 4.083 milliseconds for the 33B-parameter MiniMax H3 on 200 GPUs, about 245 chunks per second, with equivalent frame throughput approaching 1,000 frames per second and a steady-state output interval shorter than that of the 1.3B-parameter Wan2.1.
An NVIDIA research team proposes the JCA model, which can simultaneously generate video and replayable 3D motion for any given skeleton, and output a camera that registers the motion to the scene. The model keeps 3D motion in a separate model that supports arbitrary skeletons, connects to a pretrained video model through a small number of cross-attention layers, and the motion model first learns skeleton movement from motion-only data.
In its annual shareholder meeting materials, Tesla confirmed in writing for the first time that its third-generation humanoid robot Optimus is ready for large-scale mass production, but did not announce a specific timeline.
Tsinghua-affiliated embodied intelligence company Qianjue Technology released the predictive world model Polibrain at the 2026 Physical AI and World Models Conference. It focuses on learning physical laws and cross-embodiment reuse; rather than generating complete future frames, it predicts state changes in a task-relevant latent representation space, forming a closed loop of perception, decision-making, execution, and feedback.
It turns out that getting a few important details right is all it takes: extracting policies using only advantage-based supervised learning can achieve state-of-the-art results in reinforcement learning (RL)!
Estimating empowerment from offline data is useful because it provides an unsupervised objective for learning the relative "value" of each state in reaching diverse outcomes. But making it computable is notoriously difficult. In our new paper, we propose a computable estimator, along with some applications!
CondenseVLA proposes adding a learnable history compression module to the decoding side of VLA policies. It uses learnable queries and multi-head cross-attention to distill the per-frame features cached by the VLM backbone into a fixed number of slot tokens, and applies auxiliary regularization that penalizes overlap among slot attention maps and decorrelates slot embeddings to reduce redundancy.
VLA-Rolling is a lightweight inference framework that reuses action tails through three stages: action-tail inheritance, per-action confidence estimation, and confidence-guided refinement based on renoising and denoising step allocation, while the underlying VLA model remains frozen.
Thinker-WM is a unified multimodal world model that jointly models and generates RGB, metric depth, and robot actions through a shared Unified Dynamics Transformer; the same set of parameters supports joint video-action generation, world prediction, policy inference, action-conditioned generation, video-to-action generation, and inverse dynamics.
FERPO (Forward Entropy-Regularized Policy Optimization) is an online maximum-entropy reinforcement learning algorithm that improves the policy by differentiating the critic's value rather than its actions, avoiding policy-update problems caused when value predictions are accurate but action derivatives are unreliable.
The EMMA framework enhances vision-language-action (VLA) policies through a generative data engine and training pipeline, addressing the high cost of collecting robot manipulation data across varied appearances. Its core component, DreamTransfer, is based on a diffusion Transformer and generates multi-view-consistent, geometrically plausible manipulation videos, supporting text-driven editing of foreground, background, and lighting while preserving 3D structure.
The paper proposes PhysWord, a reasoning-before-acting robotic foundation model that uses atomic action words as shared semantic anchors to align continuous representations of physical dynamics with language. PhysWord predicts upcoming atomic action words and aligns their contextual hidden states with implicit targets derived from future observations, constructing a temporally structured PhysWord space with visual and optional tactile variants.
The paper proposes InteractAgent, a multimodal large language model agent system for long-horizon human-scene interaction that completes planning, execution, observation, and reflection in a closed loop, with a single MLLM serving as both planner and evaluator and a frozen MotionLLM as the executor.
The paper proposes the FIBER framework, which uses the executable latent space of a frozen behavioral foundation model (BFM) as a shared planning representation for heterogeneous interaction skills. Its latent policy projection (LPP) maps independently trained task policies into this representation, enabling low-frequency planning while retaining the BFM's high-frequency state feedback; a single perceptive flow planner is then trained entirely offline. The paper also proposes an adaptive-horizon BFM (AH-BFM) trained with multi-gamma forward-backward learning.
The paper proposes an end-to-end pipeline that uses an instrument-state logger mounted on the shaft of a standard laparoscopic instrument (combining inertial, time-of-flight, and Hall sensors, with no external camera or tracker) to capture the instrument pose and jaw state during a surgeon's operation, which is then used to train a surgical robot policy and validated on live animals.
The paper proposes WorldScape Policy 2.0, a controllable World Action Model (WAM) with reasoning-augmented long short-term memory. Its causal short-term visual memory supplies recent observations as DiT prefill to preserve local interaction dynamics, while its long-term event memory organizes historical VLM outputs into global-history, local-active, and event-boundary representations for progress-aware retrieval and to generate implicit subgoal conditions.
The paper proposes a modular method that autonomously collects data in local segments of contact-rich tasks and uses offline reinforcement learning to train local policies, which are then integrated with traditional motion planning to complete long-horizon assembly tasks. Across four real-world tasks, using only 2-2.5 hours of autonomous data collection, the method achieved an average success rate of 96% on long-horizon tasks, compared with 55% for a planning-only baseline and 37% for the strongest end-to-end baseline; the local policies also reduced applied force by 49%. Performance remained at 90% in scene-level out-of-distribution scenarios, with more than 1,500 real-world evaluation rollouts completed in total.
The paper proposes KITE (Kinematic Interaction Transfer across Embodiments), which splits manipulation into embodiment-agnostic interaction intent prediction and embodiment-specific kinematic decoding, connected through an interaction intent representation based on contact patterns.
DEFLECT is an offline post-training framework that turns the temporal gap between prediction and execution into preference supervision, enabling policies to produce actions aligned with the execution-time state even when their inputs are stale.
Researchers propose an interpretability pipeline to analyze what a fine-tuned vision-language-action (VLA) policy is computing internally. On pi0.5, injecting task A's activations into the action expert while the policy views task B's scene caused 99.6% of episodes (n=1968) to execute task A's motor trajectory, indicating the injected program is bound to absolute workspace coordinates and ignores the visible scene.
Why it matters: The study uses activation injection to break down the internal computation of VLA policies, offering a mechanistic explanation for understanding their perturbation brittleness.
A study on deep reinforcement learning for legged robots proposes using latent actions learned from a small set of expert demonstrations as an inductive bias to help joint-level torque control policies explore more efficiently. This prior allows the agent to not be limited by the demonstration reward level and significantly improves performance on transfer tasks; when combined with style rewards for imitation, it can more closely match expert behavior. Related videos have been made public.
The paper proposes a model that can simultaneously co-generate video and 3D motion for any given skeleton, along with a camera that registers that motion to the scene. The model keeps 3D motion in a separate module to support arbitrary skeletons, connects it to a pretrained video model through a small number of cross-attention layers, and uses a generative video editor to curate nearly 200K training pairs. Evaluations show that its text-driven animation of diverse skeletons is more faithful than video-first pipelines, that the motion responds to the surrounding scene, and that it also supports video-to-skeleton motion, skeleton motion-to-video, and keyframe-based 3D motion control.