object slips from a human-hand gripper even contact is detected

[qiuyuchen14]

Hi! I'm trying to use a customized gripper (a human hand model from MANO dataset) to grasp some ycb objects. I converted all the collision meshes into convex meshes using vhacd, and could detect contact between fingers and the object. However, the object always seems to manage to slip away:

Issues:
(1) I tried different friction for plane, object, and hand, from 0.5 to 10, 15, ..50, and different forces 10, 20, ...1000, and different object mass like 0.1, 0.2, ....1, but none of the combinations works, am I missing something or did anything wrong? (2) I suspect it might be the issue from hand collision mesh, even I converted it to vhacd, the force at each mesh face might not be vertical to the object surface. I replace some finger meshes with iCub robot meshes (which seems working here: https://github.com/robotology-playground/pybullet-robot-envs) but this still couldn't solve the problem:

Is there anything wrong with my code? Thank you so much!

Code:
def load_obj(obj_path, obj_q, obj_t, gt_hand_q, gt_hand_t, gt_tips, table_r, table_t):

p.resetDebugVisualizerCamera(1.8000001907348633, 18.799999237060547, 40.600013732910156, [-0.15737509727478027, 0.7639374732971191, 0.5210656523704529])

sim_timestep = 1/240
p.setTimeStep(sim_timestep)


plane_q = Rot.from_matrix(table_r).as_quat()

plane=p.loadURDF("plane.urdf", [table_t[0], table_t[1], table_t[2]], [plane_q[0], plane_q[1], plane_q[2], plane_q[3]])
p.changeDynamics(plane, -1, lateralFriction=1000)
texUid = p.loadTexture("table_surface.jpg")
p.changeVisualShape(plane, -1, textureUniqueId=texUid)
initid = 0

gt_q = gt_hand_q[initid]
gt_t = gt_hand_t[initid][0]

hand = p.loadURDF("human_hand_test_vhacd1.urdf", gt_t, gt_q, globalScaling=1, useFixedBase=1)

for objid in range(len(obj_path)):

    obj = p.loadURDF(obj_path[objid], obj_t, obj_q)
    p.changeDynamics(obj, -1, lateralFriction=10)

p.setGravity(-9.8*table_r[0, 2], -9.8*table_r[1,2], -9.8*table_r[2,2])
force_value = 100

for handid in range(len(gt_hand_q)):

    gt_q = gt_hand_q[handid]
    gt_t = gt_hand_t[handid][0]

    pos = np.array(p.calculateInverseKinematics2(hand, [31, 4, 7, 10, 13, 16, 19, 22, 25], gt_tips[handid][:].tolist()))

    for j in range(p.getNumJoints(hand)):
        p.setJointMotorControl2(hand, j, p.POSITION_CONTROL, targetPosition=pos[j], positionGain=1, velocityGain=1, force=force_value)

    p.resetBasePositionAndOrientation(hand, gt_t, gt_q)

    for _ in range(60):
        p.stepSimulation()
        time.sleep(sim_timestep)