Re: Isometric friction for planar pushing

2020-01-20T18:42:37+00:00

Update with a simpler example and minimum reproducing code (depends on numpy and matplotlib).
This time I'm pushing the block directly in block frame by applying an external force without any pusher.
However we still see that the dynamics is strongly correlated with yaw:

Code:

import loggingimport mathimport pybullet as pimport timeimport numpy as npimport pybullet_datafrom matplotlib import pyplot as pltlogger = logging.getLogger(__name__)logging.basicConfig(level=logging.DEBUG,                    format='[%(levelname)s %(asctime)s %(pathname)s:%(lineno)d] %(message)s',                    datefmt='%m-%d %H:%M:%S')logging.getLogger('matplotlib.font_manager').disabled = Truedef rotate_wrt_origin(xy, theta):    return (xy[0] * math.cos(theta) - xy[1] * math.sin(theta),            xy[0] * math.sin(theta) + xy[1] * math.cos(theta))def angular_diff(a, b):    """Angle difference from b to a (a - b)"""    d = a - b    if d > math.pi:        d -= 2 * math.pi    elif d < -math.pi:        d += 2 * math.pi    return ddef get_dx(px, cx):    dpos = cx[:2] - px[:2]    dyaw = angular_diff(cx[2], px[2])    dx = np.r_[dpos, dyaw]    return dxdef dx_to_dz(px, dx):    dz = np.zeros_like(dx)    # dyaw is the same    dz[2] = dx[2]    dz[:2] = rotate_wrt_origin(dx[:2], px[2])    return dzinit_block_pos = [0.0, 0.0]init_block_yaw = -0.physics_client = p.connect(p.GUI)p.setTimeStep(1. / 240.)p.setRealTimeSimulation(False)p.setAdditionalSearchPath(pybullet_data.getDataPath())blockId = p.loadURDF("./block_big.urdf", tuple(init_block_pos) + (-0.02,),                     p.getQuaternionFromEuler([0, 0, init_block_yaw]))planeId = p.loadURDF("plane.urdf", [0, 0, -0.05], useFixedBase=True)p.resetDebugVisualizerCamera(cameraDistance=0.5, cameraYaw=0, cameraPitch=-85,                             cameraTargetPosition=[0, 0, 1])STATIC_VELOCITY_THRESHOLD = 1e-6def _observe_block(blockId):    blockPose = p.getBasePositionAndOrientation(blockId)    xb = blockPose[0][0]    yb = blockPose[0][1]    roll, pitch, yaw = p.getEulerFromQuaternion(blockPose[1])    return np.array((xb, yb, yaw))def _static_environment():    v, va = p.getBaseVelocity(blockId)    if (np.linalg.norm(v) > STATIC_VELOCITY_THRESHOLD) or (            np.linalg.norm(va) > STATIC_VELOCITY_THRESHOLD):        return False    return Truep.setGravity(0, 0, -10)# p.changeDynamics(blockId, -1, lateralFriction=0.1)p.changeDynamics(planeId, -1, lateralFriction=0.5, spinningFriction=0.3, rollingFriction=0.1)F = 200MAX_ALONG = 0.075 + 0.2for _ in range(100):    p.stepSimulation()N = 100yaws = np.zeros(N)z_os = np.zeros((N, 3))for simTime in range(N):    # observe difference from pushing    px = _observe_block(blockId)    yaws[simTime] = px[2]    p.applyExternalForce(blockId, -1, [F, F, 0], [-MAX_ALONG, MAX_ALONG, 0.025], p.LINK_FRAME)    p.stepSimulation()    while not _static_environment():        for _ in range(100):            p.stepSimulation()    cx = _observe_block(blockId)    # difference in world frame    dx = get_dx(px, cx)    dz = dx_to_dz(px, dx)    z_os[simTime] = dz    logger.info("dx %s dz %s", dx, dz)    time.sleep(0.1)logger.info(z_os.std(0) / np.abs(np.mean(z_os, 0)))plt.scatter(yaws, z_os[:, 2])plt.xlabel('yaw')plt.ylabel('dyaw')plt.show()

You'll need to place this urdf file in the same directory (save as block_big.urdf):

Code:

Statistics: Posted by zhsh — Mon Jan 20, 2020 6:42 pm

Isometric friction for planar pushing

2020-01-20T19:59:57+00:00

Update: see next post for minimum reproducable for a simplified experiment!

I'm trying to do quasi-static planar pushing by position controlling the pusher (brown) with p.createConstraint.
The issue I'm having is that the dynamics seem to be different with different yaw of the pusher-block system (keeping their relative positions the same). I'm investigating this by telling the pusher to push perpendicular to the adjacent face for a fixed position for different yaws of the whole system.
Control of the pusher is done by

Code:

p.changeConstraint(self.pusherConstraint, newPos, maxForce=200)

then waiting for velocities to die down for quasi-staticness. I've also tried manually interpolating to the goal position by controlling to intermediate waypoints in case the previous method causes the block to leave contact (it doesn't).
The configuration of the system:

Code:

block dimensions: 0.15 x 0.15 x 0.05mass: 6lateral_friction 1.0spinning_friction: 0.1pusher dimensions: 0.15 x 0.02mass: 10lateral_friction: 1.0spinning_friction: 0.1

The test I'm doing is to push 0.02 into the block, for 11 different system yaws from 0 to 3.1 (radians). I then look at the change in body frame (block x, block y, block yaw, pusher position along block). I summarize the trials with the standard deviation along each dimension relative to the absolute value of that dimension's mean. For example I get

Code:

tensor([0.0285, 0.2691, 0.2714, 0.2669])

For trials that result in block frame differences of:

Code:

tensor([[ 0.0184, -0.0022, -0.0309, -0.0007],        [ 0.0181, -0.0028, -0.0378, -0.0008],        [ 0.0172, -0.0039, -0.0531, -0.0011],        [ 0.0183, -0.0023, -0.0307, -0.0006],        [ 0.0187, -0.0019, -0.0260, -0.0005],        [ 0.0184, -0.0023, -0.0304, -0.0006],        [ 0.0176, -0.0033, -0.0455, -0.0010],        [ 0.0176, -0.0034, -0.0462, -0.0010],        [ 0.0187, -0.0018, -0.0247, -0.0005],        [ 0.0184, -0.0022, -0.0301, -0.0006]])

These should be the same if friction is isometric right? I've tried making the block bigger, but it doesn't really help.

Statistics: Posted by zhsh — Sat Jan 18, 2020 6:18 pm

Real-Time Physics Simulation Forum

Re: Isometric friction for planar pushing

Isometric friction for planar pushing