Introduction

The RoboWorld Challenge 2026 advances embodied world modeling for robotics and autonomous driving. It brings together world models, vision-language-action (VLA) models, and embodied policies to study how agents ground language in visual observations, anticipate future states, and act safely.

Three tracks cover complementary settings: WorldNav for language-conditioned navigation without intermediate feedback, HA-VLN for human-aware social navigation with continuous observations, and SafeDrive-VLA for language-guided driving under instruction-scene conflicts.

The competition combines offline trajectory evaluation and closed-loop simulation, with track-specific datasets, metrics, and reference baselines. World-model and VLA approaches are encouraged, alongside policy-based methods where permitted by each track.

Track #1WorldNav

Open-loop visual navigation from one initial RGB observation and a natural-language instruction.

Track Details

Track #2HA-VLN 2.0

Closed-loop RGB-D navigation with socially grounded language and dynamic, interacting humans.

Track Details

Track #3SafeDrive-VLA

Language-guided driving that follows safe instructions and responds safely to instruction-scene conflicts.

Track Details

RoboPAD Workshop

RoboWorld Challenge 2026 is an independently organized challenge associated with RoboPAD: Post-Training Adaptation of Robot Foundation Models, a workshop at NeurIPS 2026.

RoboPAD focuses on adaptation, reasoning, evaluation, and safety for robot foundation models after pretraining. The challenge complements these themes through language-conditioned navigation, social interaction, and safe autonomous driving.

Challenge logistics, registration, and evaluation are managed by the RoboWorld organizing team. Challenge participation is separate from RoboPAD paper submission. Workshop program and attendance details are available on the workshop website.

Visit RoboPAD Workshop

Competition Tracks

Track #1 · Open-loop navigation

WorldNav Language-Conditioned World Navigation

Track #1

Participants develop world-model-based or vision-language-action (VLA) agents for language-conditioned visual navigation. Given one initial egocentric RGB observation and an instruction, agents generate a full trajectory of continuous planar displacement and yaw actions, ending with Stop or a null action.

No goal image or intermediate feedback is provided. The track encourages planning grounded in predicted future observations, alongside VLA and policy-only approaches that map vision and language to actions.

Track #2 · Closed-loop social navigation

HA-VLN 2.0 Human-Aware Social Navigation

Track #2

Participants develop agents that follow language instructions referencing human activities in continuous 3D environments. Using egocentric RGB-D observations, agents must reach the goal while maintaining socially compliant distances and avoiding collisions with moving people.

HA-VLN 2.0 uses 0.25 m forward steps and 15-degree turns. The track welcomes policy-based, VLA, and world-model approaches that handle partial observability, anticipate human motion, and re-plan as people block corridors or doorways.

Track #3 · Safety-aware driving

SafeDrive-VLA Towards Safety in Autonomous Driving

Track #3

Participants develop vision-language-action models (VLAs) for safe navigation guided by natural-language instructions. Given onboard visual observations and an instruction, models generate driving trajectories or control actions that account for surrounding traffic conditions.

Models should follow safe instructions, including turns, lane changes, and target-speed requests. When an instruction conflicts with the traffic scene, models must respond appropriately by slowing down, waiting, or selecting a safe alternative.

Timeline

Registration Opens

Register for your team by filling in this Google Form.
Release of Training Data and Baselines

Training data and reference baselines will be released to help teams get started across all three tracks.
Competition Servers Online @ CodaBench

Evaluation servers and public leaderboards will open on CodaBench for all three tracks.
Competition Deadline

Submit your final entries for official evaluation before the competition closes.
Award Decision Announcement

Final results and award-winning teams will be announced for each track.