Hardware Setup›Assembly

ABC Hardware Setup

#Shopping list

#Bill of materials

For The main mount

For Mounting the robot

For Mounting the wrist camera and gripper

For Cage building

For Wiring and electronics

For The robot workspace

#3D printed parts

#Tools

• Allen keys
• Electric screwdriver, for the gripper and camera-mount screws
• X-acto knife or similar, to cut out the board

#Assembly

#The main mount

In this section, we will build the primary mount for the robot and the top camera. The mount should approximately replicate the simulation environment: the empty bimanual scene abc_sim/models/yam_bimanual_empty.xml, which abc_sim/scene_xml.py loads and transforms at run time. The lower half of Figure 1 is a render of that scene with the dimensions we are matching.

1. Start with the 3060 Robot Mounting Extrusion (A) and attach the 200 mm extrusion (B) to it at the midpoint, using the black 2835 corner brackets (C), M6 T-slot sliding nuts (D) and M6 bolts (E).
2. Now attach the 1 m extrusion (F) to B using the same joint as in 1.
3. Now attach the top camera mount to F using a 10 mm M6 bolt (E) and another sliding T-nut (D). The camera mount is 3D-printed (see 3D printed parts).

#Mounting the robot

Now, we will mount the robot on the mount we built in The main mount.

Robot base

1. Slide in 4 sliding T-nuts into the base rail we set up in The main mount. Then, screw in the robot base to it using M6 20 mm bolts.
2. Do this symmetrically for the robot on the other side with the same dimensions.

Clamping to the table

3. Clamp the setup above to the top of a table (preferably adjustable height and with lockable wheels). See Figure 3.

#Mounting the wrist camera and gripper

Now, we would like to add a gripper and wrist camera to our robot. Our dataset is primarily collected on crank shaft grippers, linear gear grippers and adaptive flexpoint grippers. A larger fraction of the data is with the crank shaft gripper, so for better performance we recommend using this. The camera mount comes along with the gripper from the vendor I2RT.

1. Push the gripper onto the ends of the arms. Important: sometimes (for the crank shaft gripper with YAM / YAM-Pro), this fitting is quite tight and requires forceful pushing with the ends properly aligned.
2. This step is just screwing parts together, should be fairly simple with an electronic screwdriver. Tightly screw in the gripper to the arm with the long screws that come with the robot. Then screw in the camera mount with the screws that come with it, and finally screw in the D405 with the holes on the mount.
3. Repeat 1, 2 for the other arm.

#Cage building instructions

We will need 10 extrusions to build the cage:

• 2 of 48 in (uncut, as bought — 1220 mm)
• 4 cut to 39 in
• 4 cut to 35 in

We will build the back of the cage first.

• Attach the four corners using the 3-way connectors. The 48″ extrusions form the cross beams and the 35″ form the vertical. Important: ensure that all the extra corner connectors are pointing out in the same direction.

  

  

• Attach the bottom two connectors as follows.

  

• Then attach the last two vertical 35″ struts using the black 2835 corner brackets (C), two per strut.

  

• We then bolt them together — with corner connectors — to make a frame that looks like this.

  

• Finally, use two more corner connectors and attach the remaining side of the back of the cage to complete it.

  

• You then want to cut out the project board to dimensions matching the size of the sides (so around 35 in × 41 in). Use double-sided tape to stick the boards to the cage boundary from all sides.

#Wiring and electronics

#Cameras

• Use JSER cables to connect the three cameras to the computer using USB ports.
• Prefer USB3 for avoiding intermittent packet drops. If your compute does not have USB3, disable the depth stream of the camera in the deploy/ folder in the code to reduce bandwidth requirements.
• Use zip ties to secure the cable to the robot to the extent possible.

#Robot arms

1. Connect each arm to the computer with a USB cable (a USB2 port is fine). No separate USB-CAN adapter is needed; the arm shows up as a CAN interface.
2. Power each arm from its 24 V supply into the connector on the arm base.
3. Check that both arms show up:

   ip link show type can

4. Add your station to PROFILES in deploy/robot/config.py with the two CAN interface names or serials.
5. Run:

   export ROBOT_PROFILE=<your_profile>
   sudo -E bash deploy/scripts/setup_can_names.sh

6. Verify:

   ip -details link show can_l_foll
   ip -details link show can_r_foll

   Both should show state UP and bitrate 1000000.

#The robot workspace

In this section, we set the robot workspace up. The dataset is collected on a rectangular workspace placed parallelly adjacent to the robot, with the cage as the visual background. While performance without the cage is not completely zero, it is much better with the cage on.

1. Choose a table of about 48 in × 39 in (1220 × 990 mm), the footprint of the cage. Preferably, the dimension away from the robot should be greater than the cage dimension (39 in).
2. Align the table with the base of the robot as in Figure 5 (left): the table top sits flush with the top of rail A, level with the bottom of the robot base plate.
3. Put the cage on and clamp it to the other table on which the robot is mounted on. Your workstation would probably end up looking like as in Figure 5 (right).