MJLAB / CL-V4

Independent RL teachers

One PPO policy per task. Trained through interaction in simulation, with recorded rollouts and measured success across the 24-task suite.

18 / 24Teachers certified
6Tasks in progress
>90%Target for every task
60D observations8D normalized actionsFranka + two-finger gripperUpdated 13 Sep 2026 · 09:27 UTC

Certified policies passed above 90% in each of two independent 128-episode evaluations. Videos show actual recorded RL rollouts. For tasks in progress, the card shows the latest evaluated policy and a failure example; newer training may still be running. Tool-Pull remains deferred.

Speed and smoothness are under review. Only Lift-Cube is being tuned; other teachers are paused. Compare Lift motion trials →

Axial Extract

Certified
100.0% / 100.0%Success example

Validation / confirmation · 128 episodes each

PPO · 4s episode · Franka
RL-017-axial-mechanism/model_200.pt

Cage Drag

Certified
93.0% / 93.8%Success example

Validation / confirmation · 128 episodes each

PPO · 4s episode · Franka
RL-013-R11-cage-refine-cpu/model_19092.pt

Drag Pull

Certified
95.3% / 93.8%Success example

Validation / confirmation · 128 episodes each

PPO · 3s episode · Franka
RL-004-R2-drag-extend/model_1300.pt

Flip Switch

Certified
100.0% / 100.0%Success example

Validation / confirmation · 128 episodes each

PPO · 3s episode · Franka
RL-009-flip-mechanism/model_800.pt

Lift Cube

Certified
92.2% / 91.4%Success example

Validation / confirmation · 128 episodes each

PPO · 20s episode · Franka
RL-002-R16-lift-guard/model_21990.pt

Open Door

Certified
100.0% / 100.0%Success example

Validation / confirmation · 128 episodes each

PPO · 3s episode · Franka
RL-010-door-mechanism/model_900.pt

Open Drawer

Certified
100.0% / 100.0%Success example

Validation / confirmation · 128 episodes each

PPO · 3s episode · Franka
RL-008-drawer-mechanism/model_1000.pt

Open Lid

Certified
100.0% / 100.0%Success example

Validation / confirmation · 128 episodes each

PPO · 5s episode · Franka
RL-015-V3-lid-bounded/model_700.pt

Place In Container

Certified
91.4% / 93.0%Success example

Validation / confirmation · 128 episodes each

PPO · 20s episode · Franka
RL-019-R8-place-box/model_7500.pt

Push Button

Certified
100.0% / 100.0%Success example

Validation / confirmation · 128 episodes each

PPO · 3s episode · Franka
RL-007-button-mechanism/model_700.pt

Push Cuboid

Certified
98.4% / 97.7%Success example

Validation / confirmation · 128 episodes each

PPO · 3s episode · Franka
RL-003-R3-push-extend/model_1998.pt

Push Flap

Certified
100.0% / 100.0%Success example

Validation / confirmation · 128 episodes each

PPO · 3s episode · Franka
RL-011-flap-mechanism/model_900.pt

Reach Target

Certified
100.0% / 100.0%Success example

Validation / confirmation · 128 episodes each

PPO · 20s episode · Franka
RL-001-reach-s20260912/model_499.pt

Rotate Valve

Certified
99.2% / 100.0%Success example

Validation / confirmation · 128 episodes each

PPO · 8s episode · Franka
RL-016-valve-mechanism/model_1999.pt

Slide Window

Certified
100.0% / 100.0%Success example

Validation / confirmation · 128 episodes each

PPO · 3s episode · Franka
RL-014-window-mechanism/model_300.pt

Throw To Bin

Certified
97.7% / 98.4%Success example

Validation / confirmation · 128 episodes each

PPO · 5s episode · Franka
RL-024-R7-throw-box/model_14098.pt

Topple Block

Certified
96.1% / 92.2%Success example

Validation / confirmation · 128 episodes each

PPO · 4s episode · Franka
RL-006-topple-resume-wandb/model_499.pt

Turn Lever

Certified
100.0% / 100.0%Success example

Validation / confirmation · 128 episodes each

PPO · 3s episode · Franka
RL-012-lever-mechanism/model_1000.pt

Edge Grasp

In progress
0.0%Failure example

0/128 passed · latest reviewed checkpoint

PPO · 6s episode · Franka
RL-021-R6-edge-capture/model_12994.pt

Peg Insertion

In progress
0.0%Failure example

0/128 passed · latest reviewed checkpoint

PPO · 20s episode · Franka
RL-020-R10-peg-guard/model_10198.pt

Pivot Lift

In progress
0.0%Failure example

0/128 passed · latest reviewed checkpoint

PPO · 6s episode · Franka
RL-022-R4-pivot-box-guard/model_8497.pt

Reorient Object

In progress
85.2%Failure example

109/128 passed · latest reviewed checkpoint

PPO · 20s episode · Franka
RL-005-R15-reorient-settle/model_20592.pt

Stack Cube

In progress
66.4%Failure example

85/128 passed · latest reviewed checkpoint

PPO · 20s episode · Franka
RL-018-R11-stack-box-guard/model_14194.pt

Strike Slide

In progress
39.1%Failure example

50/128 passed · latest reviewed checkpoint

PPO · 4s episode · Franka
RL-023-R11-strike-returns/model_19490.pt