task stringclasses 1
value | instance_id int64 301 310 | rollout_id int64 0 0 | steps int64 3.23k 3.23k | success bool 1
class | agent_distance dict | normalized_agent_distance dict | q_score dict | time dict |
|---|---|---|---|---|---|---|---|---|
turning_on_radio | 301 | 0 | 3,225 | false | {
"base": 1.82112409624375,
"left": 2.71997917150793,
"right": 3.584989712184324
} | {
"base": 3.1274639722507303,
"left": 1.1977665248788842,
"right": 1.1359865235202131
} | {
"final": 0
} | {
"simulator_steps": 3225,
"simulator_time": 107.5,
"normalized_time": 0.6665550387596899
} |
turning_on_radio | 302 | 0 | 3,225 | false | {
"base": 3.5207733453505625,
"left": 7.172786790792998,
"right": 5.79266119687145
} | {
"base": 1.6176843668511982,
"left": 0.45420282172360765,
"right": 0.7030447425786805
} | {
"final": 0
} | {
"simulator_steps": 3225,
"simulator_time": 107.5,
"normalized_time": 0.6665550387596899
} |
turning_on_radio | 303 | 0 | 3,225 | false | {
"base": 1.909684678364556,
"left": 2.49378503712137,
"right": 3.123320914748547
} | {
"base": 2.9824295416548017,
"left": 1.306407710169223,
"right": 1.3039005952828482
} | {
"final": 0
} | {
"simulator_steps": 3225,
"simulator_time": 107.5,
"normalized_time": 0.6665550387596899
} |
turning_on_radio | 304 | 0 | 3,225 | false | {
"base": 0.6502320762502336,
"left": 1.2264275790880674,
"right": 2.088574021116983
} | {
"base": 8.75918031119732,
"left": 2.6564144965024927,
"right": 1.949894980414436
} | {
"final": 0
} | {
"simulator_steps": 3225,
"simulator_time": 107.5,
"normalized_time": 0.6665550387596899
} |
turning_on_radio | 305 | 0 | 3,225 | false | {
"base": 1.23718925696312,
"left": 2.30019179230203,
"right": 3.13556159228483
} | {
"base": 4.603580226666792,
"left": 1.4163601534894168,
"right": 1.2988103981183283
} | {
"final": 0
} | {
"simulator_steps": 3225,
"simulator_time": 107.5,
"normalized_time": 0.6665550387596899
} |
turning_on_radio | 306 | 0 | 3,225 | false | {
"base": 1.4150258614284894,
"left": 4.630456816108563,
"right": 4.747002288814656
} | {
"base": 4.025014775525239,
"left": 0.7035806896344063,
"right": 0.857909845461001
} | {
"final": 0
} | {
"simulator_steps": 3225,
"simulator_time": 107.5,
"normalized_time": 0.6665550387596899
} |
turning_on_radio | 307 | 0 | 3,225 | false | {
"base": 0.6411746534596432,
"left": 1.6783044661447093,
"right": 1.8751821166661102
} | {
"base": 8.882915082915838,
"left": 1.941185324665097,
"right": 2.171789056542681
} | {
"final": 0
} | {
"simulator_steps": 3225,
"simulator_time": 107.5,
"normalized_time": 0.6665550387596899
} |
turning_on_radio | 308 | 0 | 3,225 | false | {
"base": 3.4733290980107654,
"left": 4.936715458293293,
"right": 5.200978202512488
} | {
"base": 1.639781270154305,
"left": 0.6599327077940019,
"right": 0.7830257773494719
} | {
"final": 0
} | {
"simulator_steps": 3225,
"simulator_time": 107.5,
"normalized_time": 0.6665550387596899
} |
turning_on_radio | 309 | 0 | 3,225 | false | {
"base": 0.8472085527682225,
"left": 1.6960056957145753,
"right": 2.3581937217682025
} | {
"base": 6.722665843481118,
"left": 1.9209251526878595,
"right": 1.7269573582557032
} | {
"final": 0
} | {
"simulator_steps": 3225,
"simulator_time": 107.5,
"normalized_time": 0.6665550387596899
} |
turning_on_radio | 310 | 0 | 3,225 | false | {
"base": 5.067125537226758,
"left": 7.543282585421366,
"right": 7.463094873713999
} | {
"base": 1.1240100443844838,
"left": 0.43189420031756826,
"right": 0.5456851438863359
} | {
"final": 0
} | {
"simulator_steps": 3225,
"simulator_time": 107.5,
"normalized_time": 0.6665550387596899
} |
BEHAVIOR-1K π0.5 baseline: turning_on_radio
The 2026 BEHAVIOR Challenge's provided π0.5 checkpoint for turning_on_radio, evaluated with the official
evaluator on public test instance indices 0-9 (instance ids 301-310). These are the 10
instances used for leaderboard reporting. Each instance ran once, with the default timeout of
1.5× the mean human demonstration length (3225 steps at 30 Hz).
Result: 0/10 successes, mean q-score 0.000.
The q-score is the fraction of BDDL goal predicates satisfied at the end of the episode.
| Instance | Success | q-score | Steps | Sim time (s) | Base distance (normalized) |
|---|---|---|---|---|---|
| 301 | no | 0.00 | 3225 | 107.5 | 3.13 |
| 302 | no | 0.00 | 3225 | 107.5 | 1.62 |
| 303 | no | 0.00 | 3225 | 107.5 | 2.98 |
| 304 | no | 0.00 | 3225 | 107.5 | 8.76 |
| 305 | no | 0.00 | 3225 | 107.5 | 4.60 |
| 306 | no | 0.00 | 3225 | 107.5 | 4.03 |
| 307 | no | 0.00 | 3225 | 107.5 | 8.88 |
| 308 | no | 0.00 | 3225 | 107.5 | 1.64 |
| 309 | no | 0.00 | 3225 | 107.5 | 6.72 |
| 310 | no | 0.00 | 3225 | 107.5 | 1.12 |
Files
results.csv one row per rollout (the table above)
turning_on_radio/json/ the evaluator's per-rollout result JSONs
turning_on_radio/videos/ rollout videos: head camera and both wrist cameras
turning_on_radio/r1pro_openpi.yaml the robot config passed with --robot-config
Setup
- Simulator: BEHAVIOR-1K
v3.9.3-post2(a8247a8cc), Isaac Sim 5.1.0, torch 2.7.0+cu128, on an RTX PRO 6000 Blackwell. - Policy: the provided π0.5
turning_on_radiocheckpoint (see the BEHAVIOR-1K docs,docs/challenge/baselines.md), served by the OpenPI challenge fork (behaviorbranch,0cc8e355f) with--control_mode receding_horizon --action_horizon 16. - Observations: the default
DefaultWrapper(224×224 RGB + proprioception).
r1pro_openpi.yaml is the evaluator's bundled r1pro.yaml with only the robot renamed from
robot_r1 to robot. The OpenPI R1Pro config reads observation keys under that name
(robot::proprio), and the default config makes the policy server fail with KeyError.
# policy server (OpenPI)
uv run scripts/b1k/serve_b1k.py --robot b1k/R1Pro --task b1k/turning_on_radio --repo-id turning_on_radio \
--policy.config pi05_b1k --policy.dir <checkpoint> \
--control_mode receding_horizon --action_horizon 16 --port 8000
# evaluator (BEHAVIOR-1K)
python -m omnigibson.eval.eval --task-name turning_on_radio --robot-config r1pro_openpi.yaml \
--instance-indices 0 1 2 3 4 5 6 7 8 9 --num-envs 10 --write-video --output-dir <dir>
License
CC BY-NC 4.0. The rendered frames show BEHAVIOR-1K assets, which are licensed for non-commercial research use only.
- Downloads last month
- 22