CLARU
mulliganMIT

sim-square-narrow-c01-dagger-mulligan-no-cf

A simulated square navigation task dataset collected as part of the Mulligan project, containing 100 episodes of training data without counterfactual augmentation.

Downloads272
Episodes100

Why This Matters for Physical AI

This dataset contributes to understanding how dagger-based imitation learning and model ablations (absence of counterfactual augmentation) affect navigation policy training in simulated environments.

Technical Profile

Modalities
proprioception
Environment
simulation
Task Types
navigation
Episodes
100
Data Format
parquet
Annotation Types
reward_labels
License
MIT
Part of the Mulligan family

Access

View on HuggingFace

Need custom proprioception data?

Claru builds purpose-built datasets for simulation applications with dense human annotations and quality assurance.

Request a Sample Pack

Related Datasets

Mulligan paper evidence

Frozen inputs behind the figures and appendix tables of the Mulligan paper, including evaluation summaries, collection ledgers, value-learning studies, simulation critic studies, and reset-range photo composites.

2K downloadsOct 2026MIT

Thread Nut: round 5 evaluation

Real-robot held-out evaluation dataset for the Mulligan paper containing episodes from evaluation sessions with multiple policy comparisons (HG-DAgger, HG-DAgger+Mulligan, HiL-IDQL+Mulligan) on thread nut tasks.

rgb
991 downloadsOct 2026MIT

real-square-d2-r02-eval-b01-mulligan-no-cf-dp-policy-rollouts

Policy rollout dataset from the Mulligan project containing 50 episodes of real robot square manipulation task demonstrations with multi-camera observations.

rgb
530 downloadsOct 2026MIT

real-square-d2-r04-eval-b01-mulligan-no-cf-dp-policy-rollouts

Policy rollout dataset from Mulligan project containing 50 episodes of real robot square manipulation task evaluations using model round R4 variant.

rgb
465 downloadsOct 2026MIT

real-square-d2-r01-eval-b01-mulligan-no-cf-dp-policy-rollouts

Policy rollout evaluation dataset for the real-square-d2 task, part of the Mulligan framework. Contains 50 episodes of robot manipulation trajectories with multi-camera observations.

rgb
458 downloadsOct 2026MIT

real-routing-d2-c01-dagger-mulligan-no-cf

Part of the Mulligan project, this dataset contains training data for a real cable routing task collected through dagger methodology with the mulligan-no-cf variant. It includes 50 episodes with multi-view camera observations.

rgb
351 downloadsOct 2026MIT