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.
DragMesh: Interactive 3D Generation Made Easy, proposed by a Peking University team, has been accepted to IJCV. It targets local interaction on existing 3D assets, recovering the target part's joint type, motion axis, and joint origin through a single drag, and generating a complete, continuous joint motion trajectory without retraining a motion model for each new object.
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!
This work presents the first fully polynomial-time approximation scheme (FPTAS) for the multi-robot motion planning problem, for a fixed number k of translating polygonal robots in a polygonal environment, covering the min-sum, min-max, and min-makespan objectives. The algorithm generalizes to polytope environments and polytope robots in any fixed dimension d, and it resolves the open problem raised by Agarwal et al. on an FPTAS for k=2 unit-square and unit-disk robots, while being simpler than existing results. The paper will be published at SODA 2027.
The paper proposes a human-directed policy search workflow for multi-agent pathfinding in which coding agents iteratively propose, implement, evaluate, diagnose, and refine algorithms in code space, combining simulation feedback with explicit correctness verification.
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.
PUN (Planning Under Noise) is a neural operator that learns noise-aware value fields to enable motion planning with control execution noise in unseen geometric environments. It changes obstacle boundary conditions to reflecting Neumann conditions and sets a Dirichlet condition at the goal, and extends the Monte Carlo solver Walk on Stars (WoSt) to higher dimensions, learning the mapping from geometry to value field from noisy, partially converged labels.
CLOSER is an evidence-aware active verification framework for object-goal navigation (ObjectNav) that treats stopping decisions as closed-loop evidence acquisition, combining vision-language model (VLM) verification with active perception.
A study proposes a distributed platoon controller that analyzes the asymptotic behavior of multi-vehicle traffic at an unsignalized intersection through closed-loop path modeling. It shows that under mild conditions on the controller gains, safe crossing is guaranteed during the transient phase, and in steady state all vehicles cross smoothly at a common constant speed while avoiding collisions. Unlike traditional bi-level decoupled designs on open paths, the method couples intersection control with platoon convergence analysis. Its effectiveness was validated in a high-fidelity CARLA simulator using multiple commercial vehicle models.
A controlled evaluation on the HighwayEnv testbed (1Hz, with gpt-4o-mini used for the key comparisons) shows that the published gains from experience memory in closed-loop LLM driving agents mainly come from sampling luck in the warmup pool: retrieval-based memory beat No-Memory in only one of ten pools (sign test p=0.011).
The paper proposes Latent Projection for Latent Planning, noting that planning costs in latent world models are often taken as the ℓ2 distance between the predicted terminal latent and the goal latent, while the latent itself contains noise whose source is precisely the regularizer used during training to prevent collapse.
An arXiv paper proposes MAPLE (Mondrian-Anchored Prior for Learning-free Indoor Exploration), a training-free predictive exploration method that adapts the Mondrian process into an online generative prior conditioned on geometric cues from the current partial map.
Idealworks and NVIDIA systematically evaluated the sim-to-real transfer of the X-Mobility and COMPASS learned navigation pipeline on industrial mobile robots. After training in a digital twin with Cosmos-generated sim-to-real data, real-robot success rates across four scenarios rose from 65.3% to 91.7%, weighted travel time nearly halved, the gap to the production-line navigation stack was 8.3 percentage points, and everything ran on the onboard Jetson AGX Xavier.
The paper proposes SPAR (Search-trained Push Action Ranker), which learns to rank push primitives from simulated search results to alleviate the exponential growth of candidate action sequences with sequence length in Navigation Among Movable Obstacles (NAMO).
The authors propose Generative Plan Latents (GPL), which represent the reasoning state of vision-language-action (VLA) models as a low-dimensional conditional distribution over plans, such as which object, grasp, or route to use, so that plans can be sampled, selected on request, and reweighted by deployment outcomes without retraining the VLA.
Proposes a self-supervised method, the Probabilistic Reach-Avoid Value Function (PVF), for computing reach-avoid sets for robot tasks under stochastic, potentially unbounded disturbances, applicable to black-box systems. Compared with Hamilton-Jacobi reachability (HJR)-based baselines on systems of increasing complexity, up to the Hopper locomotion task, results show that when disturbances are non-adversarial, PVF's probabilistic reach-avoid sets are less conservative and more accurate than adversarial HJR sets. The method has been submitted to ICRA 2027.
The study examines whether expected free energy (EFE) can determine how much computation to allocate to planning based on a learned world model, with an objective combining goal distance and disagreement among stochastic predictions. Experiments found that stopping when elite cost stalls reduces refinement while mean success rates stay close to fixed-depth references; with a frozen model, rewarding disagreement lowers success, while online updates partially recover losses caused by regions missing from the training data. The cost-stall rule fails when optimization exploits model errors, and predicted cost keeps falling after task success rates peak.
A study proposes a high-precision remote center of motion (RCM) servo architecture for ophthalmic robots, using a hierarchical QP controller and authority-guided recovery planning to address servo oscillation caused by conflicts between RCM and tip-tracking objectives. In a 200-scenario kinematic benchmark, the approach achieved a 100% reach rate for 128 near-singular targets and a 99% reach rate for 72 near-joint-limit targets, with software RCM residual below 0.142 mm.
An agent-centered framework incorporating physical constraints discovers interpretable car-following laws from multi-source trajectory data, yielding the Adaptive Equilibrium Response Model (AERM), which couples state-dependent relative-speed response with spacing-dependent equilibrium-speed restoration.
An arXiv preprint proposes a posture-aware navigation framework for bipedal humanoid robots, whose core components include the Adaptive Traversal Tubes (ATT) corridor width field, Interval Velocity–Heading (IVH) control, and a sidestep-enhanced AMP locomotion prior.
CovNav is a training-free framework that uses conformal prediction (CP) to turn spatial queries in ObjectNav into structured binary decisions: a local large model in fast mode judges whether the target is likely near an observed object, CP converts its token probabilities into a prediction set, and only ambiguous queries are escalated to slow chain-of-thought reasoning.
An arXiv paper studies how an omnidirectional evader escapes in minimum time from the forward conic field of view of a Reeds-Shepp pursuer with minimum turning radius r. For slower or equally fast evaders with speed ratio sinα<ν≤1, it constructs a feedback solution using Isaacs' characteristics and gives a scalar equation for the finite escape time.
This study compares 10 deterministic methods and 2 LLM-based variants for heterogeneous multi-robot task allocation, covering rule-based, similarity-based, requirement-oriented, fuzzy, and semantic definitions of suitability. The results show that the methods often disagree on scoring and ranking candidate robots in the same scenario, that generic similarity measures can violate the suitability property of one-sided requirement satisfaction, and that LLM scoring introduces stochasticity and linguistic sensitivity. The paper was accepted to the FM-MRS Workshop at IROS 2026.
The paper proposes an autonomous bricklaying concept system based on a Boston Dynamics Spot quadruped robot with a manipulator arm, operating between a supply area and an assembly area, using AprilTag localization and autowalk mission navigation, and exporting target brick poses directly from CAD software for the robot to execute.
A method for autism spectrum disorder (ASD) education enables a camera-equipped Nao robot, using a vision-based large language model, to convert photos of a Shoebox Task into a Planning Domain Definition Language (PDDL) representation and generate action plans, autonomously executing the task, prompting the next step, or reporting failures through action demonstrations, voice, and gestures.
The paper proposes LaLa-World, a language-conditioned latent world model (LWM) that learns jointly from simulation and robot trajectories as well as broader image and video data. Its key design is to describe both numerical actions and visual changes in language, enabling training of a single future predictor across heterogeneous datasets, and to build the predictor on a language model so it can interpret language conditions.
CausGround proposes a non-parametric causal interaction grounding framework that distills interaction experience into action-centered precondition and effect knowledge, and combines it with current observations during planning for backward and forward reasoning.
The "space-air-ground integrated desert planting robot" project from Fudan University's Qi Lizhe team won the innovation gold award in the university exhibition area at the 26th China International Industry Fair. The robot integrates autonomous perception, precise navigation, and integrated planting functions, combining BeiDou, inertial navigation, LiDAR, and stereo vision positioning, and has completed field operation verification in the desert of Dalad Banner, Ordos, Inner Mongolia. The team has filed 7 core invention patents, and the single-vehicle platform will later integrate four planting mechanisms, with operating efficiency reaching more than 10 times that of traditional manual work.
REEMAN has launched the Waterproof Ant (model WBOT10F), an outdoor waterproof autonomous forklift with a rated load of 2 tons. It features a waterproof body and chassis protection design, aimed at material handling in outdoor factory areas on rainy days.
AdaExploreBench introduces a benchmark that instantiates environmental dynamics as hidden rules, requiring LLM agents to actively acquire evidence and infer the rules, and, under implicit rule drift, to detect that prior knowledge has become stale and re-explore. Experiments on eight leading LLMs show that GPT-6 Sol achieves only 61.9% calibrated accuracy; providing sufficient evidence or removing rule drift improves average performance by close to 10%. The results indicate that agents can reason from valid evidence but struggle to actively acquire evidence and to recognize when evidence becomes invalid.
ScaleNav proposes a unified world-action framework that achieves language-conditioned local navigation through three-stage training: first, instruction-conditioned video pretraining on internet first-person videos to learn how observations evolve as the camera approaches a target; then, aligning predicted executable action chunks using action-labeled simulation data; and finally, jointly fine-tuning all trainable components.
Agiwei Robotics has introduced a new three-dimensional embodied intelligence paradigm based on VLA (Vision-Language-Agent) technology, used for environmental understanding and autonomous decision-making by unmanned forklifts under complex industrial conditions. Its self-developed fully dynamic laser SLAM navigation fuses LiDAR and 3D vision, achieving repeat positioning accuracy of ±5 mm; the fourth-generation MCU controller coordinates with high-precision servo motors for motion control. Agiwei's mobile robot solutions have entered industries including photovoltaic new energy, petrochemicals, and automotive parts, with cumulative delivery to 900 customers and annual capacity of 3,000 units (sets).
A TMTPost interview and analysis of Niuwa Robotics founder Yang Ruigang notes that after a decade of unmanned delivery development, China's logistics unmanned vehicle fleet exceeds 35,000 units and more than 200 cities have opened road rights, yet the last few hundred meters—entering residential compounds, boarding elevators, and delivering to the doorstep—remain unsolved. Most explorations at home and abroad, including JD.com, Cainiao, and Amazon Scout, have largely exited. The core reasons are ambiguous rules for private spaces, high deployment costs, and the need for human fallback for exceptions; no company has ever made this account profitable.
Dianchetong tested Robotaxi services from WeRide, Pony.ai, and DiDi Autonomous Driving in Guangzhou after a two-year gap. All three have removed safety drivers, and their operating areas, driving efficiency, and ride experience have improved markedly compared with two years ago.
Metro-WM proposes a hierarchical planning framework that, instead of generating unconstrained latent sub-goals, retrieves real states from observed frames in offline expert demonstrations or random-action trajectories as sub-goals, and connects frames from different episodes into paths toward the goal. Experiments show its long-horizon success rate is up to 37.33 percentage points higher than the next-best hierarchical method, up to 10.9 times faster, uses 13-56 times less offline computation, and requires less tuning. The authors also note that leading macro planners frequently output physically unrealizable sub-goals.
NEXUS is a modular neuro-symbolic framework that jointly maintains a PDDL action domain model and an LTL safety specification: execution failures are used to repair action schemas, while detected hazards are converted into reusable LTL constraints through hierarchical semantic and translation sampling with equivalence consensus.
An ECCV 2026 accepted paper proposes UniWM, a memory-augmented world model that unifies egocentric visual foresight and planning within a single multimodal autoregressive backbone, grounding action selection in visually imagined outcomes and tightly aligning prediction with control.
The paper proposes using Conditional Flow Matching to reformulate long-horizon goal prediction in open-environment active 3D mapping as conditional multimodal anchor generation, replacing traditional single-point prediction.
The paper proposes OptCar, a method that specializes a generalist forward kinodynamic (FKD) model into a vehicle-specific model while preserving cross-terrain generalization. At its core, a FiLM mechanism summarizes recent state-action history into a single dynamics context token that conditions multi-step predictions in a Transformer FKD architecture; specialization then combines limited real-vehicle data with targeted synthetic trajectories generated by system identification.