Vision-and-Language Navigation (VLN) requires embodied agents to ground language in visual observations and make long-horizon decisions. Existing methods often address these demands through task-specific architectures. As multimodal large language models (MLLMs) become increasingly capable, an important question remains: Have we truly unlocked the potential of native MLLMs for VLN before introducing task-specific architecture? We present SimpleVLN, an architecture-preserving framework developed through studies of navigation formulation, data construction, and optimization. At the formulation level, SimpleVLN combines single-turn decision inputs with fixed-budget global-local memory to efficiently use navigation history without unbounded context growth, while action chunks enable efficient output of multiple primitive actions per model invocation. For data construction, we combine expert demonstrations with policy-dependent DAgger experience and develop a scalable synthesis pipeline that generates large-scale tasks for reinforcement learning. At the optimization level, a progressive IL-DAgger-RL pipeline advances from action imitation to task-level outcome optimization. Comprehensive experiments validate the effectiveness of these three aspects. We instantiate SimpleVLN across multiple MLLM backbones and model sizes, demonstrating its generality. SimpleVLN achieves 76.7% and 80.1% success rates on the R2R-CE and RxR-CE splits, respectively, substantially outperforming existing methods. These results show that the navigation capabilities of native MLLMs can be effectively unlocked without task-specific architectural designs.
Figures & tables
Figure 1: Motivation and performance comparison of SimpleVLN . (a) Motivation for using native MLLMs in VLN. (b) Comparison with existing methods on R2R-CE and RxR-CE.
Figure 2: Overview of SimpleVLN. At each decision, sampled history frames, current observation, and the instruction are fed to an MLLM, which directly predicts an action chunk for navigation.
Figure 3: Training data pipeline of SimpleVLN. SimpleVLN first learns from expert demonstrations, then collects and aggregates corrective trajectories through DAgger. For RL, it synthesizes navigation episodes from sampled start–goal paths and generated instructions, then optimizes the policy through environment interaction and outcome feedback.
Method
Size
Observation
R2R-CE val-unseen
RxR-CE val-unseen
Pano.
Odo.
Depth
S.RGB
NE ↓
OS ↑
SR ↑
SPL ↑
NE ↓
SR ↑
SPL ↑
nDTW ↑
NaVid ( Zhang et al., 2024b )
7B
✓
5.47
49.1
37.4
35.9
8.41
23.8
21.2
—
Uni-NaVid ( Zhang et al., 2024a )
7B
✓
5.58
53.3
47.0
42.7
6.24
48.7
40.9
—
NaVILA ( Cheng et al., 2024 )
7B
✓
5.22
62.5
54.0
49.0
6.77
49.3
44.0
58.8
ActiveVLN ( Zhang et al., 2025 )
3B
✓
5.31
58.3
50.1
43.7
5.84
50.7
41.2
58.1
StreamVLN ( Wei et al., 2025 )
7B
✓
4.98
64.2
56.9
51.9
6.22
52.9
46.0
61.9
Table 1: Comparison with state-of-the-art methods on the R2R-CE and RxR-CE val-unseen splits. † and ∗ denote SimpleVLN before and after RL, respectively.
Method
Size
Latency (ms) ↓
R2R-CE
RxR-CE
NE ↓
OS ↑
SR ↑
SPL ↑
NE ↓
SR ↑
SPL ↑
nDTW ↑
StreamVLN ( Wei et al., 2025 )
7B
300
4.98
64.2
56.9
51.9
6.22
52.9
46.0
61.9
JanusVLN ( Zeng et al., 2026 )
7B
906
4.78
65.2
60.5
56.8
6.06
56.2
47.5
62.1
DualVLN ( Wei et al., 2026 )
7B
683
4.05
70.7
64.3
58.5
4.58
61.4
51.8
70.0
FutureNav ( Zhang et al., 2026c )
4B
1141
4.24
70.2
65.4
61.3
4.60
60.9
52.8
68.0
SimpleVLN
4B
293
3.39
79.4
73.8
66.7
3.29
75.1
65.3
71.5
Table 2: Comparison with state-of-the-art methods on inference latency and navigation performance. Latency is measured on R2R-CE using the same NVIDIA A100 GPU, and measures model execution only for one action-chunk decision, excluding preprocessing, environment execution, and other non-model overhead. SimpleVLN uses Qwen3-VL-4B as its backbone.
Table 3: Navigation interface and memory design. Ablations of action chunking, single-turn navigation, and observation sampling using Qwen3-VL 4B fine-tuned on R2R. The default design of SimpleVLN is shaded.
Figure 4: Memory budget. Effect of global and local memory sizes on R2R-CE val-unseen SR. We vary M with N=1 (left) and N with M=8 (right); stars indicate the default configuration.
R2R
RxR
ScaleVLN
DAgger
R2R-CE
RxR-CE
NE ↓
OS ↑
SR ↑
SPL ↑
NE ↓
SR ↑
SPL ↑
nDTW ↑
✓
5.53
61.2
53.6
48.1
6.45
50.3
42.0
57.6
✓
✓
5.28
63.4
56.7
51.3
5.75
54.5
46.7
62.4
✓
✓
✓
4.90
68.4
57.9
50.9
5.27
55.3
48.1
63.6
✓
✓
✓
✓
4.26
74.3
65.4
58.8
4.79
63.4
54.1
66.5
Table 4: Demonstration and DAgger data ablation on the R2R-CE and RxR-CE val-unseen splits.
R2R
RxR
ScaleVLN
R2R-CE
RxR-CE
NE ↓
OS ↑
SR ↑
SPL ↑
NE ↓
SR ↑
SPL ↑
nDTW ↑
Pre-RL baseline
4.26
74.3
65.4
58.8
4.79
63.4
54.1
66.5
✓
3.91
74.0
68.5
63.2
4.31
65.8
55.3
68.3
✓
3.76
71.8
67.5
59.3
3.98
68.5
59.5
68.8
✓
4.17
70.9
66.6
61.3
4.34
66.3
56.2
67.3
✓
✓
3.75
74.1
69.4
63.8
3.77
69.7
60.8
68.1
Table 5: Synthesized RL data ablation on the R2R-CE and RxR-CE val-unseen splits. Dataset columns identify the training-scene sources used to synthesize RL episodes.
RL configuration
R2R-CE
RxR-CE
NE ↓
OS ↑
SR ↑
SPL ↑
NE ↓
SR ↑
SPL ↑
nDTW ↑
Pre-RL baseline
4.26
74.3
65.4
58.8
4.79
63.4
54.1
66.5
GRPO + SPL(τ)
3.80
73.9
68.9
63.1
4.13
67.7
58.3
67.3
GRPO + I(success)
3.74
73.1
70.1
65.6
3.82
71.0
62.1
68.4
DAPO + I(success)
3.45
76.6
71.5
65.3
3.39
73.7
64.0
69.4
Table 6: Reward design and RL optimization on the R2R-CE and RxR-CE val-unseen splits with Qwen3-VL 4B. The SPL reward is zero for failed episodes and equals the ratio of the ground-truth path length to the executed path length for successful episodes; the success reward is binary.
Figure 5: RL training dynamics. Overall, rollout reward and evaluation success rate increase, while the average number of action-chunk decisions per trajectory decreases.
Figure 6: SimpleVLN in real-world and simulated environments. SimpleVLN follows diverse instructions and reaches the specified goals; arrows indicate navigation directions.