VLA Gallery / All models
211 Robot VA/VLA Models
Spec pages for 211 robot VA/VLA models (2022–2026): backbone LLM, action head, training GPUs and data, inference device and memory, RL usage and base control.
Reference baselines (pre-2024)8 models
| RT-1 | 2022-12 | A 35M real-robot policy conditioned on language instructions via FiLM, without a pretrain… |
| RT-2 | 2023-07 | Co-fine-tunes a pretrained VLM on robot data and web VQA·caption data, outputting actions… |
| RT-1-X | 2023-10 | A cross-embodiment model training the RT-1 architecture on mixed OXE data from 9 embodime… |
| RT-2-X | 2023-10 | Co-fine-tunes RT-2 on original VLM data and the OXE robot mix at about 1:1. |
| ACT | 2023-04 | Low-cost bimanual imitation learning policy that regresses action chunks at once and aver… |
| Diffusion Policy (CNN) | 2023-03 | Policy that generates action sequences by diffusion denoising to handle multimodal action… |
| Diffusion Policy (Transformer) | 2023-03 | A variant implementing the same diffusion policy as the CNN variant with a Transformer de… |
| GR-1 | 2023-12 | Pretrained by future-frame prediction on large-scale human video, then jointly learns act… |
VA17 models
| Mobile ALOHA | 2024-01 | Low-cost whole-body teleoperated mobile bimanual system; co-trains with static ALOHA data… |
| ALOHA Unleashed | 2024-10 | Trains a per-task diffusion transformer on thousands of teleoperation demonstrations per… |
| HPT | 2024-09 | Pretrains on heterogeneous robot·sim·human video·deployed robot data with a shared trunk… |
| BAKU | 2024-06 | Trains an about 10M-parameter multitask policy on one consumer GPU in 6–26 h per benchmar… |
| VQ-BeT | 2024-03 | Converts continuous actions into hierarchical residual VQ codes and predicts them in one… |
| ScaleDP | 2024-09† | Reports performance gains from scaling a diffusion policy transformer from 10M to 1B. |
| UMI | 2024-02 | Collects demonstrations without a robot using a handheld gripper with a GoPro and transfe… |
| DP3 | 2024-03 | 3D diffusion policy that encodes sparse point clouds with a small MLP and learns from abo… |
| Data Scaling Laws | 2024-10 | Collected over 40k demonstrations with UMI and showed that generalization follows a power… |
| RISE | 2024-04 | Encodes a single-view point cloud with sparse 3D convolutions and outputs actions via a t… |
| Robot Utility Models (RUM) | 2024-09 | Uses a $25-BoM iPhone collection tool to gather about 1,000 demos per task across about 4… |
| EgoMimic | 2024-10 | Co-trains one policy on first-person human demos captured with Project Aria glasses and r… |
| InterACT | 2024-09† | An ACT-family policy that models inter-arm dependencies with hierarchical attention and p… |
| CARP | 2024-12† | Converts action sequences into multi-scale tokens and predicts them autoregressively from… |
| Dense Policy | 2025-03 | Generates action sequences in logarithmic time by bidirectional dense expansion with an e… |
| Chain-of-Action | 2025-06 | Generates the trajectory backward from the goal keyframe to the current time, reflecting… |
| BRS (WB-VIMA) | 2025-03 | Learns long-horizon household whole-body manipulation with Galaxea R1, which has a mobile… |
Autoregressive & latent-action VLA24 models
| Octo | 2023-12 | A generalist VA policy trained on OXE 800k episodes without a pretrained VLM. |
| OpenVLA | 2024-06 | An open 7B VLA trained on 970k OXE trajectories; ECoT·TraceVLA·OpenVLA-OFT·UniVLA and oth… |
| OpenVLA-OFT | 2025-02 | Changes the OpenVLA fine-tuning recipe to parallel decoding·action chunks·L1 regression,… |
| MiniVLA | 2024-12 | Shrinks OpenVLA from 7B to about 1B and adds VQ action chunks and history input. |
| TinyVLA | 2024-09 | Attaches a diffusion head to a small VLM without robot-data pretraining and trains with L… |
| ECoT | 2024-07 | Generates concrete reasoning such as plans·subtasks·object locations before actions, rais… |
| LLaRA | 2024-06 | Converts robot actions to text in LLaVA conversation format and trains with 6 kinds of au… |
| 3D-VLA | 2024-03 | Combines 3D scene understanding and goal state generation (diffusion world model) with ac… |
| RoboFlamingo | 2023-11 | A continuous-action VLA that freezes the VLM and trains only the resampler·cross-attentio… |
| SpatialVLA | 2025-01 | Reduces 7-dim actions to 3 tokens with 3D positional encoding and adaptive grid tokens. |
| LAPA | 2024-10 | Pretrains a VLA by learning latent actions from robot·human videos without action labels. |
| UniVLA (OpenDriveLab) | 2025-05 | Learns latent actions in two stages, task-agnostic and task-centric, so even human video… |
| Magma | 2025-02 | A multimodal agent foundation model that handles both UI navigation and robot manipulatio… |
| TraceVLA | 2024-12 | Improves spatiotemporal awareness with visual trace prompting, feeding point-tracking tra… |
| CoT-VLA | 2025-03 | Visual chain-of-thought (visual CoT) VLA that first generates a subgoal image. |
| NORA | 2025-04 | A 3B autoregressive VLA adding FAST+ action tokens to Qwen2.5-VL-3B. |
| MolmoAct | 2025-08 | An 'action reasoning model' that outputs actions via depth perception and visual trajecto… |
| WorldVLA | 2025-06 | Unifies action generation and next-image prediction (world model) in one autoregressive m… |
| UniVLA (BAAI) | 2025-06 | Treats vision·language·actions as one discrete token sequence and performs world-model po… |
| Moto | 2024-12 | Pretrains a GPT on latent motion tokens learned from video, then connects them to real ac… |
| villa-X | 2025-07 | Uses latent actions as an intermediate plan, trained jointly with a robot action expert. |
| VLA-0 | 2025-10 | A design that makes the VLM output actions as plain-text numbers without changing its arc… |
| VLA-Adapter | 2025-09 | Reached LIBERO 97.3% with a 0.5B backbone and Bridge Attention policy without robot pretr… |
| RoboMamba | 2024-06 | VLA that uses a Mamba state-space model as the LLM instead of a transformer. |
π family9 models
| π0 | 2024-10 | The first π-family model, attaching a separate flow matching action expert to a VLM backb… |
| π0-FAST (FAST 토크나이저) | 2025-01 | Compresses action chunks with DCT and BPE so high-frequency continuous control can be lea… |
| Hi Robot | 2025-02 | A hierarchical system in which a high-level VLM turns user prompts and mid-task intervent… |
| π0.5 | 2025-04 | One model first predicts subtask text and then outputs actions, co-trained with web, mult… |
| π0.5 + KI | 2025-05 | Blocks action-expert gradients from flowing into the VLM to preserve VLM knowledge while… |
| Real-Time Chunking (추론시 RTC) | 2025-06 | Inference technique applied to diffusion/flow VLAs without retraining; inpaints the next… |
| π0.6 | 2025-11 | Scaled the backbone to Gemma 3 4B and the action expert to about 860M, and added rich met… |
| π*0.6 (RECAP) | 2025-11 | Iteratively trains on demonstrations·autonomous rollouts·expert corrections with advantag… |
| 학습시 RTC | 2025-12 | Simulates inference latency during training (uniformly sampled 0–10 steps on hardware) to… |
Flow·Diffusion VLA18 models
| RDT-1B | 2024-10 | Diffusion policy trained jointly on 46 heterogeneous robot datasets through a unified act… |
| RDT2 | 2025-09 | Trained on about 10,000 hours of human demonstrations collected in 100+ homes with an imp… |
| CogACT | 2024-11 | A separate DiT generates action sequences conditioned on the cognition token from the VLM… |
| DiVLA (Diffusion-VLA) | 2024-12 | Injects VLM-generated reasoning sentences into the diffusion policy head via FiLM, with t… |
| DexVLA | 2025-02 | Uses a 3-stage curriculum that first pretrains a 1B diffusion expert on cross-embodiment… |
| HybridVLA | 2025-03 | One LLM does both autoregressive prediction and diffusion action generation, using a coll… |
| Dita | 2024-10† | Performs diffusion denoising with in-context conditioning that puts language·image·timest… |
| SmolVLA | 2025-06 | 0.45B small VLA pretrained on 481 community SO100 datasets, aimed at single-GPU training… |
| X-VLA | 2025-10 | Absorbs heterogeneity across 7 platforms with per-embodiment soft prompts (0.04% of all p… |
| FLOWER | 2025-09 | Pretrained in about 200 H100 GPU-hours, cutting parameters by over 20% via VLM mid-layer… |
| InternVLA-M1 | 2025-10 | A dual system that first pretrains the VLM on spatial grounding to locate key objects, th… |
| WALL-OSS | 2025-09 | Uses Uni-CoT, which produces reasoning (CoT), subtask plans, and continuous actions in on… |
| EO-1 | 2025-08 | Jointly learns reasoning and actions in one decoder from interleaved sequences mixing vis… |
| GraspVLA | 2025-05 | Without real-robot data, co-trains synthetic SynGrasp-1B (10M trajectories) and web groun… |
| Evo-1 | 2025-11† | Reports Meta-World 80.6% and RoboTwin 37.8% at 0.77B without robot-data pretraining, high… |
| NORA-1.5 | 2025-11 | Attaches a flow expert to an autoregressive VLA, jointly learning FAST+ tokens and contin… |
| F1-VLA | 2025-09 | A VLA that generates future visual scenes (foresight) in a next-scale manner before actio… |
| DexGraspVLA | 2025-02 | An off-the-shelf VLM planner specifies the target object, and a DiT controller taking seg… |
Industry models24 models
| Gemini Robotics | 2025-03 | Splits into a cloud backbone and an onboard decoder, taking about 250ms from observation… |
| Gemini Robotics-ER | 2025-03 | VLM that outputs spatial understanding results such as pointing, 3D bboxes, and grasp pre… |
| Gemini Robotics On-Device | 2025-06 | Designed for minimal compute so it runs locally on the robot without a data network. |
| Gemini Robotics 1.5 | 2025-09 | A 'Thinking VLA' that reasons in natural language before acting, training on data from mu… |
| Gemini Robotics-ER 1.5 | 2025-09 | A high-level 'brain' that orchestrates robot activity, issuing step-by-step natural-langu… |
| RT-H | 2024-03 | Hierarchical VLA in which one model first predicts a language motion such as 'move arm fo… |
| AutoRT | 2024-01 | A system that orchestrates a robot fleet with LLMs·VLMs to collect real-world data at sca… |
| SARA-RT | 2023-12 | An efficiency technique that replaces quadratic-complexity attention in pretrained·fine-t… |
| GR-2 | 2024-10 | Pretrains future-frame generation on 38M web video clips, then jointly fine-tunes video·a… |
| GR-3 | 2025-07 | Co-trains on robot trajectories, web vision-language data and VR human trajectories to ge… |
| GR-RL | 2025-12 | Filters demonstrations with a learned task progress (Q-value), then after left-right mirr… |
| GR-Dexter | 2025-12 | Controls a fixed bimanual setup with two 21-DoF humanoid hands (56-DoF total) using a 4B… |
| TRI LBM 1.0 | 2025-07 | A diffusion transformer policy conditioned on CLIP language without an LLM·VLM. |
| GO-1 | 2025-03 | Plans with embodiment-agnostic latent actions learned from human video, and an Action Exp… |
| Genie Envisioner | 2025-08 | A lightweight action decoder (GE-Act) takes representations from a video world model (GE-… |
| RFM-1 | 2024-03 | Handles language instruction understanding, in-context learning, and action-conditioned v… |
| Skild Brain | 2025-07 | Positions itself as an omni-bodied model controlling quadrupeds·humanoids·tabletop arms·m… |
| DYNA-1 | 2025-06 | Finds and fixes its own mistakes with a reward model that evaluates task progress, aiming… |
| GEN-0 | 2025-11 | An embodied foundation model pretrained on 270,000 h of real-world manipulation data. |
| Galaxea G0 | 2025-08 | A two-tier design where a VLM planner handles high-level instructions and a VLA executor… |
| GigaBrain-0 | 2025-10 | Reduces dependence on real-robot data with generated data from the GigaWorld world model… |
| RynnVLA-001 | 2025-08 | Three-stage training: pretrains video generation on 12M egocentric human manipulation vid… |
| UnifoLM-WMA-0 | 2025-09 | Unitree's open framework combining a world model and action generation, demonstrated with… |
| V-JEPA 2-AC | 2025-06 | Trains a world model self-supervised on over 1 million hours of internet video, then atta… |
Humanoid VLA14 models
| GR00T N1 | 2025-03 | Attaches pseudo-actions from latent actions (LAPA) and an inverse dynamics model (IDM) to… |
| GR00T N1.5 | 2025-05 | Freezes the VLM in both pre- and post-training and adds a FLARE (future latent representa… |
| GR00T N1.6 | 2025-09 | Removes N1.5's 4-layer adapter after the VLM, unfreezes the top 4 VLM layers, and doubles… |
| Figure Helix | 2025-02 | Projects the S2 latent vector into the S1 token space, concatenates it with visual featur… |
| Atlas LBM | 2025-08 | A single policy directly outputs whole-body actions including gripper joints·neck yaw·tor… |
| 1X Redwood AI | 2025-06 | Predicts locomotion, manipulation, and pelvis pose commands together with arm and hand co… |
| Tesla Optimus | 2024-05(2차) | No official technical disclosure; per secondary sources, an end-to-end neural network usi… |
| Being-H0 | 2025-07 | Uses 'physical instruction tuning', chaining VLA pretraining on human hand videos, physic… |
| Humanoid-VLA | 2025-02 | A hierarchical structure that generates whole-body motion from language and egocentric vi… |
| LeVERB | 2025-06 | Hierarchical VLA that links a vision-language layer and a whole-body control layer throug… |
| UH-1 (Humanoid-X) | 2024-12 | A VLA-adjacent model that converts language into humanoid motion without vision input. |
| EgoVLA | 2025-07 | Pretrained on human egocentric video, then fine-tuned on robot demonstrations, transferri… |
| HAT | 2025-03 | Co-trains human VR demonstrations and humanoid teleoperation in the same 54-D action spac… |
| WholeBodyVLA | 2025-12 | Large-space loco-manipulation through unified latent learning from low-cost egocentric vi… |
Humanoid whole-body controllers18 models
| HumanPlus HST | 2024-06 | A low-level tracker for shadowing that transfers human motion to a humanoid in real time. |
| HumanPlus HIT | 2024-06 | A visuomotor policy that learns autonomous whole-body skills from up to 40 demonstrations… |
| Humanoid Locomotion as Next Token Prediction | 2024-02 | Treats real-world locomotion as next-token prediction and jointly trains on trajectories… |
| Challenging-Terrain Transformer | 2024-10 | Pretrained via sequence modeling on trajectories from existing policies·model-based contr… |
| BeyondMimic | 2025-08 | Adaptive sampling learns the hard segments of long motions, transferring LAFAN1 motions t… |
| LeVERB-A | 2025-06 | Distills vision-language-agnostic RL teachers into a transformer student via DAgger, conv… |
| OmniH2O | 2024-06 | A tracker for whole-body teleoperation via VR·RGB keypoints, and a platform for training… |
| HOVER | 2024-10 | Unifies multiple control modes into one multi-mode whole-body policy via command modes·sp… |
| iDP3 | 2024-10 | An egocentric 3D point cloud diffusion policy that performs humanoid upper-body manipulat… |
| ExBody2 | 2024-12 | Automatically filters training motions with a tracking-error threshold τ* = 0.150 to trac… |
| TWIST | 2025-05 | Performs mocap-based whole-body teleoperation with a PPO privileged teacher and an RL+BC… |
| TWIST2 | 2025-11 | Collects teleoperation data without mocap using PICO4U VR and a custom 2-DoF robot neck (… |
| AMO | 2025-05 | Adapts to torso commands using hybrid data generated by trajectory optimization (Crocoddy… |
| GMT | 2025-06 | Adaptive sampling (motions over 10 s cut to 10 s and weighted by completion) and MoE let… |
| SONIC | 2025-11 | A general token-based whole-body controller that scales motion tracking in data·model·com… |
| VIRAL | 2025-11 | Distills an RL teacher into an RGB student in large-scale tiled-rendering sim, transferri… |
| BumbleBee | 2025-06 | A specialist-generalist training framework combining motion clustering and sim-to-real ad… |
| Figure RL 보행 제어기 | 2025-03 | Simulates thousands of Figure 02 units with different physical parameters in parallel, le… |
New in 202635 models
| G0Plus | 2026-01 | Galaxea's January 2026 public checkpoint (G0Plus_3B-base), pretrained on over 2k hours of… |
| InternVLA-A1 | 2026-01 | Coordinates scene understanding, visual prediction generation, and action execution withi… |
| 1X World Model (1XWM) 정책 | 2026-01 | A video-generation-based policy in which an IDM extracts actions between video frames gen… |
| Spirit v1.5 | 2026-01 | Trained on diverse, loosely controlled collected data instead of curated data. |
| Being-H0.5 | 2026-01 | Trained 30 embodiments on UniHand-2.0 (16,000+ hours of human egocentric·14,000+ hours of… |
| Cosmos Policy | 2026-01 | A video model policy that trains a pretrained video model as a policy without architectur… |
| Rho-alpha (ρα) | 2026-01(2차)† | The first robotics model derived from the Microsoft Phi family, a 'VLA+' that adds tactil… |
| LingBot-VLA | 2026-01 | Pretrained on about 20,000 h of real data from 9 bimanual robot types. |
| Helix 02 | 2026-01 | S1 expanded from upper body to whole-body control, and S0 is a learned whole-body control… |
| UnifoLM-VLA-0 | 2026-01 | Performs 12 categories of manipulation tasks on Unitree G1 with a single policy. |
| HoloBrain-0 | 2026-02 | Injects embodiment information via multi-view camera parameters and kinematic description… |
| G0Tiny | 2026-02 | Galaxea's 250M lightweight model, using a SmolVLM2 backbone. |
| Xiaomi-Robotics-0 | 2026-02 | Learns asynchronous execution with a Λ-shaped attention mask and RoPE position offsets. |
| DreamZero | 2026-02 | A world action model (WAM) that jointly predicts video and actions. |
| EgoScale | 2026-02 | Pretrains a VLA on over 20,854 hours of egocentric human video, raising average success o… |
| Ψ0 (Psi-Zero) | 2026-03 | Reported beating baselines trained on over 10× more data by 40%+ using only about 800 h o… |
| GR00T N1.7 | 2026-03 | An open reasoning VLA pretrained on 20K hours of EgoScale human video and robot demonstra… |
| GEN-1 | 2026-04 | Pretraining data contain no robot data; trained on over 500,000 h of human wearable-devic… |
| GO-2 | 2026-04(2차) | Combines Action CoT, which plans directly in action space, with an asynchronous dual syst… |
| Being-H0.7 | 2026-04 | A 3B model introduced by BeingBeyond as its 'flagship WAM', reporting a LIBERO average of… |
| π0.7 | 2026-04 | Its key feature is diverse context conditioning that adds execution quality·speed·strateg… |
| MolmoAct 2 | 2026-05 | Action reasoning model; the Think variant uses adaptive-depth reasoning tokens. |
| G0.5 | 2026-06 | Decoder-only VLA that produces reasoning and actions in a single autoregressive stream. |
| SyVLA | 2026-06 | An academic VLA that disclosed its pretraining compute (32 H200s, about 1.5 days). |
| OpenHLM | 2026-06 | A humanoid recipe that fine-tunes π0.5 to output whole-body commands and stacks it on the… |
| Galbot AstraBrain(银河星脑) 계열 | 2026-06(2차) | A family consisting of 银河星脑/AstraBrain, AstraBrain-Agent, AstraBrain-WBC 0.5, and DexNDM… |
| InternVLA-A1.5 | 2026-07† | Supervises latent visual prediction tokens with a WAN2.2-5B video teacher during training… |
| TurboVLA | 2026-07 | A 0.2B lightweight VLA reporting 31.2 ms inference latency (32 Hz) and 0.9 GB inference V… |
| Gemini Robotics 2 | 2026-07 | A VLA that converts visual·language input into whole-body motor control from toes to fing… |
| Gemini Robotics On-Device 2 | 2026-07 | A VLA that runs locally on the robot, adapting to new bimanual embodiments in a few hours… |
| DYNA-2 | 2026-08 | A world action model (WAM) pretrained on over 1M hours of egocentric human video. |
| Skild S1 | 2026-08(Skild)/2026-09(NVIDIA) | An in-context learning VA that performs new tasks from a single video demonstration witho… |
| GEN-1.5 | 2026-08 | Takes a 3–12 second demonstration as a physical prompt and performs new tasks without wei… |
| ω-0 (OMEGA-0) | 2026-08 | A humanoid latent-predictive world action model that predicts future observation embeddin… |
| Qwen-VLA | 2026-05 | Attaches a DiT flow-matching decoder to a Qwen VLM, unifying manipulation·navigation·traj… |
Mobile manipulation16 models
| TidyBot++ | 2024-11 | Learns a per-task Diffusion Policy from phone teleoperation demonstrations on an omnidire… |
| MoManipVLA | 2025-03 | Transfers a VLA to mobile manipulation by using bi-level optimization to find base waypoi… |
| AC-DiT | 2025-07 | Feeds the latent representation of a lightweight mobility head, pretrained on base action… |
| SG-VLA | 2026-03 | Strengthens spatial grounding by co-training, in stages with the VLA, auxiliary decoders… |
| PanoVLA | 2026-08 | A Mixture-of-Transformers inserting a panorama expert into π0.5 fuses global spatial cont… |
| HarmonicMM | 2023-12 | Trains a single policy end-to-end with DD-PPO in ProcTHOR sim that moves base and arm tog… |
| ACT-1 | 2025-11 | Converts human demonstrations collected with Skill Capture Glove into robot data via Skil… |
| UMI on Legs | 2024-07 | A diffusion manipulation policy trained on UMI handheld demonstrations outputs task-frame… |
| VBC | 2024-03 | A hierarchical quadruped+arm whole-body control framework in which a visual high-level po… |
| Helpful DoggyBot | 2024-09 | Modular indoor object-fetching system for a Go2 with a front-bottom 1-DoF gripper, combin… |
| QUAR-VLA (QUART) | 2023-12 | Proposed the quadruped VLA task definition, in which quadruped high-level commands (veloc… |
| QUART-Online | 2024-12 | A quadruped VLA that removes latency without shrinking parameters by fine-tuning an MLLM,… |
| MoRE | 2025-03 | A quadruped VLA that inserts many LoRA experts into Fuyu-8B as a sparse MoE, treats the V… |
| RoboDuet | 2024-03 | A two-stage-trained whole-body control framework in which two RL policies, a locomotion p… |
| Learning Visual Quadrupedal Loco-Manipulation from Demonstrations | 2024-03 | A hierarchical leg-manipulation policy: an RL low-level controller tracks a parameterized… |
| SLIM | 2025-01 | A quadruped+arm policy that performs long-horizon search·approach·grasp·carry·drop tasks… |
Navigation28 models
| GNM | 2022-10 | Unifies heterogeneous driving data from 6 robots with a normalized action space and tempo… |
| ViNT | 2023-06 | Trains a 31M Transformer policy on multi-robot driving data that extends the GNM data mix… |
| NoMaD | 2023-10 | Handles goal-directed navigation and goal-free exploration in one diffusion policy via go… |
| CrossFormer | 2024-08 | Trains a single 130M Transformer on 900K trajectories from 20 embodiments without alignin… |
| LeLaN | 2024-10 | Attaches object instructions to YouTube tour·human walking videos and robot driving data… |
| CityWalker | 2024-11 | Imitation-learns a GPS-goal-conditioned urban navigation policy from thousands of hours o… |
| NWM | 2024-12 | Trains a 1B conditional diffusion transformer (CDiT) world model on robot driving video a… |
| X-Mobility | 2024-10 | An end-to-end navigation model that first learns environment dynamics with an autoregress… |
| COMPASS | 2025-02 | Trains per-embodiment residual RL experts on top of an imitation-learning-based policy, t… |
| OmniVLA | 2025-09 | An omni-modal navigation VLA taking language·goal images·2D goal positions alone or combi… |
| RING (The One RING) | 2024-12 | Builds a shape-agnostic indoor object search policy by imitation learning on expert traje… |
| NavDP | 2025-05 | Using only simulation data, jointly trains trajectory generation (diffusion) and trajecto… |
| NaVid | 2024-02 | A video-VLM-based VLN agent that outputs the next action as language from monocular RGB v… |
| Uni-NaVid | 2024-12 | Unifies the input/output format of 4 navigation tasks (VLN·ObjectNav·EQA·human following)… |
| NaVILA | 2024-12 | A two-stage design where the VLA outputs mid-level language actions with distances·angles… |
| Mobility VLA | 2024-07 | Training-free hierarchical navigation policy that feeds an entire environment demonstrati… |
| StreamVLN | 2025-07 | A streaming VLN model that processes continuous video as a multi-turn dialogue, reusing r… |
| NavFoM | 2025-09 | A navigation foundation model trained as one model on multiple embodiments (quadrupeds·dr… |
| InternVLA-N1 | 2025-07 | An asynchronous dual-system navigation foundation model where a VLM-based System 2 output… |
| TrackVLA | 2025-05 | An embodied visual tracking VLA that places a language head for target recognition and an… |
| VLN-R1 | 2025-06 | Applies SFT to a general VLM (Qwen2-VL) for egocentric-video-based VLN, then improves mul… |
| ABot-N0 | 2026-02 | Unifies 5 navigation tasks (point·object·instruction·POI·person following) in one VLA wit… |
| VLingNav | 2026-01 | Navigation VLA with adaptive CoT that turns on explicit reasoning only when needed and a… |
| TIC-VLA | 2026-02 | A think-in-control VLA in which a fast action policy controls asynchronously, conditioned… |
| FutureNav | 2026-06 | Brings world-action modeling to VLN, jointly learning action prediction, inverse·forward… |
| Uni-LaViRA | 2026-05 | A training-free agent in which a pretrained MLLM picks a direction in language and points… |
| LightNav-0 | 2026-08 | Adds only pointing tokens and RVQ action tokens to a pretrained VLM, with no new head, un… |
| Qwen-RobotNav | 2026-06 | Parameterizes task mode and observation context (token budget·temporal decay·camera weigh… |