Organizations: Shanghai Jiao Tong University, Shanghai, China. · Eastern Institute of Technology, Ningbo, China. · Department of Aeronautical and Aviation Engineering, The Hong Kong Polytechnic University, Hung Hom, Kowloon, Hong Kong. · University of Science and Technology of China, Hefei, China.
Agile and reliable navigation is a core capability for embodied robots navigating complex spatial constraints and dynamic interactions. However, a single policy often struggles to accommodate the distinct decision-making demands of diverse scenarios. We present TAME-Nav, an asymmetric Teacher-Student navigation framework built around a temporal soft mixture of experts (Soft-MoE). The Teacher combines clean observations with causally maintained dynamic-obstacle tracks to construct privileged representations, while the Student learns complementary navigation features from corrupted observation histories through temporal encoding, soft expert routing, and latent distillation. Procedural scenario generation and parallel task-level simulation broaden training coverage, while deployment-informed asymmetric domain randomization supports sim-to-real transfer. Across in-distribution and unseen compositional scenarios, TAME-Nav achieves higher success rates, lower collision rates, and higher navigation scores than representative baselines. On a real quadruped robot, it completes all twenty trials across local-trap escape and narrow-passage navigation with sudden dynamic obstacles, with only one minor contact.
Figures & tables
Fig. 1: Overview of TAME-Nav. The Teacher combines clean observation histories with causal dynamic tracks, while the temporal Soft-MoE Student learns from corrupted histories through latent distillation. Both share an Actor with LiDAR-based safety correction. A privileged Critic supports training; only the Student encoder, Actor, and safety module are deployed.
Fig. 2: Representative layouts from the six training scene families, each containing 100 distinct maps.
Fig. 3: In-distribution navigation performance. (OM) denotes training on our maps. TAME-Nav achieves the highest success rate across all six scene families and the best overall SR, CR, and Score, with particularly clear gains in dense clutter and concave traps.
Fig. 4: Mean expert routing weights across scene families. Boxed cells indicate the largest weight in each row; 12.5% corresponds to uniform routing.
Fig. 5: An example layout from the sequential compositional test set, with trajectories of our method and the OM baselines.
Method
SR (%) ↑
CR (%) ↓
Score ↑
SEA-Nav
47.27
52.73
0.1934
NavRL
0.00
75.39
0.0000
SEA-Nav (OM)
66.80
33.01
0.2393
DCLP (OM)
59.96
40.04
0.2998
NavRL (OM)
34.96
60.94
0.1743
TAME-Nav (Ours)
81.64
18.16
0.4068
TABLE I: Sequential compositional navigation performance. Our method achieves the highest SR and Score and the lowest CR. (OM) denotes training on our maps; unmarked baselines use their original maps.
Fig. 6: An example multi-room layout with paired trajectories of our method and the OM baselines. Dynamic obstacles are shown at their initial positions together with their motion ranges; each robot trajectory spans its full episode.
Method
SR (%) ↑
CR (%) ↓
Score ↑
SEA-Nav
48.24
37.70
0.2234
NavRL
3.91
42.97
0.0195
SEA-Nav (OM)
66.41
22.85
0.2535
DCLP (OM)
60.16
35.16
0.2951
NavRL (OM)
67.58
15.62
0.3115
TAME-Nav (Ours)
75.20
13.48
0.3448
TABLE II: Compositional multi-room navigation performance. Our method achieves the highest success rate and Score and the lowest collision rate in unseen multi-room environments.
Fig. 7: Real-world local-trap scenario and illustrative trajectories. Our method temporarily deviates from the goal direction to bypass the concave barrier and reach the goal.
Method
SR(%) ↑
Collisions/trial ↓
SEA-Nav
0
0.30
NavRL
0
1.00
SEA-Nav (OM)
0
0.00
DCLP (OM)
0
0.20
NavRL (OM)
50
0.80
TAME-Nav (Ours)
100
0.10
TABLE III: Real-world local-trap navigation over ten trials per method. Our method completes all trials with only one minor contact in total, whereas no baseline exceeds 50% success. Collision counts are averaged per trial. One minor collision is allowed; a second collision, obstacle toppling, or persistent entrapment constitutes failure.
Fig. 8: Narrow-passage scenario with unexpected dynamic obstacles and illustrative trajectories. Obstacles abruptly enter the robot’s path midway through the passage and near the goal. Our method negotiates both encounters and reaches the goal.
Method
SR(%) ↑
Collisions/trial ↓
SEA-Nav
80
1.00
NavRL
0
1.00
SEA-Nav (OM)
90
0.70
DCLP (OM)
80
0.70
NavRL (OM)
90
0.70
TAME-Nav (Ours)
100
0.00
TABLE IV: Real-world dynamic obstacle avoidance in narrow passages over ten trials per method. Our method completes all trials without collisions, while successful baselines incur repeated contacts. Collision counts are averaged per trial.
Fig. 9: Ablation results on the in-distribution test set. All ablations degrade performance, with removing Teacher tracks causing the largest drop in success rate.
The pursuit of general-purpose embodied agents is hindered by fragmented evaluation protocols that isolate navigation skills and fixate on specific robot morphologies, failing to reflect real-world scenarios where agents must orchestrate diverse behaviors across varying embodiments. To bridge this gap, we introduce OmniNavBench, a benchmark for cross-skill coordination and cross-embodiment generalization. OmniNavBench introduces three paradigm shifts: (1) Compositional Complexity. We propose composite instructions that interleave sub-tasks from 6 categories (PointNav, VLN, ObjectNav, SocialNav, Human Following and EQA), compelling agents to transition between exploration, interaction, and social compliance within a single episode. (2) Morphological Universality and Sensor Flexibility. We present a simulation platform that breaks the reliance on single-morphology evaluation, enabling generalization tests across humanoid, quadrupedal, and wheeled robots, with a modular sensor interface and 170 environments blending synthetic assets with real-world scans. (3) Demonstrations Quality. Moving beyond shortest-path algorithms, we curate 1779 expert trajectories via human teleoperation, capturing behavioral nuances such as exploratory glance and anticipatory avoidance. Extensive evaluations demonstrate that current methods, despite their claimed unified design, struggle with the complex, interleaved nature of general-purpose navigation. This exposes a critical disparity between existing capabilities and real-world deployment demands, underscoring OmniNavBench as a testbed for the next generation of generalist navigators. Dataset, code, and leaderboard are available at http://omninavbench.cloud-ip.cc.
Learning-based methods have demonstrated strong performance in mapless robot navigation. However, a learned policy remains restricted to the embodiment on which it was trained. Many cross-embodiment navigation methods require time-consuming online training and can still underperform target-specific policies. In fact, demonstrations collected on a source embodiment already contain navigation experience that is potentially reusable for deriving a policy for each target embodiment, but the embodiment mismatch prevents these demonstrations from directly serving as valid supervision for the target. In this article, we present CeDR-Nav, a cross-embodiment demonstration recasting framework that trains a dedicated navigation policy for each target embodiment from existing source demonstrations without online data collection. Geometry-consistent scaling with per-demonstration validity assessment converts the source demonstrations into target training data, and a target-specific Gaussian mixture policy generates feasible avoidance maneuvers for each target embodiment. Extensive quantitative simulations and real-world deployments across multiple robot platforms demonstrate the effectiveness of CeDR-Nav, with an average success-rate gain of 13.8 percentage points over DRL-DCLP in cross-embodiment simulations and target-specific policy training completed within seconds.
Xianghui Wang, Xinming Zhang, Yiwei Qian +4
Department of Mechanical Engineering, The Hong Kong Polytechnic University, Hung Hom, Kowloon, Hong Kong · College of Information Science and Technology, Eastern Institute of Technology, Ningbo, China · School of Computer Science and Technology, University of Science and Technology of China, Hefei, China +2
Language provides a natural interface for humans to specify and reason about goals, but training language-conditioned robotic policies and reasoning typically requires expensive manual annotations. We propose Action Self-Distillation (ASD), a method for learning language-conditioned reasoning for robotic navigation without human-annotated language or reasoning labels. ASD uses an open-source VLM, which contains broad world knowledge and reasoning capabilities, as both a teacher network and a student policy. During training, the self-teacher receives privileged trajectory feedback and infers the underlying navigation goal together with goal-oriented reasoning. The student is then trained to condition on the inferred navigation goal, predict future ego-frame trajectory, and provide a description of the scene semantics, plan, and reasoning, from onboard observations alone. Across 13 challenging indoor and outdoor real-world navigation tasks, our method improves success rate over the strongest baseline by 30% in indoor settings, 30% on multi-stage tasks, and achieves 95% success rate on outdoor tasks, validating that privileged trajectory feedback can enable self-distillation for scalable generalist navigation policies without human annotations.