I just arrived back from a second session of hacking and casting with Jim Bredt over at Viridis3D, and I have to say I’m pretty excited. The biggest result of the most recent round of mold making is a successful method for getting soluble cores into printed molds and casting silicone around them reliably and repeatably. I ended up doing a lot of revisions to the casting method and industrial design based on the results of last month’s experiments. The mold goes together in new ways with changes to accommodate pouring the mold, reusing the outer shell, aligning the cores, printing the cores, and how it will get hooked up to pneumatics once everything’s cast and functioning. I’m especially proud of the design behind the base of the soluble core, which is tapered so that it can align and center itself within the mold even if its dimensions shrink by a few percent. The little ears on this base are both to align it rotationally (which doesn’t matter much as far as the tentacle is concerned, but could matter in other more complex molds) and to prevent the core dropping down further into the mold and mucking up the wall thickness if it shrinks more than spec.

The current system of molding and casting involves printing an inner soluble mold out of a corn starch based material, and an outer rigid mold out of a new material developed by Viridis3D called MakerDust. The MakerDust shell gets shored up with a two part resin once it comes out of the machine and becomes a fairly rigid, durable composite. I haven’t tested it to its limits, yet, but it seems more durable than plaster and at least comparable to the Zcorp standard print material that gets infiltrated with superglue.

The outer mold has a few features that will hopefully make prototyping and iterating fairly quick. First, it conserves parts as much as possible. Powder printing has the habit of getting thrown off by small factors in the print head, the powder, the infiltration of the ink, and even the humidity in the air. This often means printing up a batch of identical parts and using the best ones out of the batch for the final product. Having part A come out fine but needing to wait until the machine sorts itself out to get a working version of part B stinks. So, the mold shell is made of two copies of the same part put together. It’s got rounded alignment pins to make sure it goes back together the same way each time and holes for 1/4″ – 20 fasteners. The whole assembly is generated by a parametric model in SolidWorks so I can adjust details about the final skin that I want and the entire assembly will reshuffle itself to compensate for the changes.

One of the major setbacks in this round of experimentation was how the starch cores have been coming out. During the first week of trials, they seemed durable enough to hold up under their own weight, plus the abuse involved in getting in all the cracks and crevices with an air tool and evicting all the loose powder that could spoil the casting. This week was a different story. The cores were falling apart at the slightest provocation. Jim ended up getting some acceptable results by designing a void into the prints in the shape of a wire running down their thickest area, pausing the print just after it rose above the level of the void, and inserting a bent wire there with some superglue. This rigidized everything enough to get the parts into an oven and get them into a usable state. Eventually we figured out that the weakness wasn’t due to interfering design goblins or adverse weather conditions. It was actually a misplaced ingredient in the mix. I was intensely relieved when I found this out, as including a wire in every core would have severely reduced the brands of geometry I could make into cores. As it stands, it seems like the “make a silicone skin with any geometry you please on the inside” idea still holds water.

So, here’s a picture of the final result directly above this sentence. The mold halves were waxed, the gaps in the surface filled in with food-grade grease, and the core was gently lowered in. The next step was mixing up some Smooth-On Dragon Skin (which is a beast to mix evenly due to the identical viscosity and color of the two parts) and slowly pouring it into the mold. I noticed the Dragon Skin holding on to a lot of bubbles that were whipped into the mix while stirring. Given how the silicone has a viscosity like honey when mixed up, I think the bubbles will stay where they are and not aggregate in spots, but I’d rather have them not be there at all. I might have to vacuum the material before casting next time.

The final result is very pretty if I do say so myself. Unfortunately I had to head back to Brooklyn before the silicone had entirely cured, but Jim sent me the mold in the post and some photos to tide me over until it arrives. From what I can tell, I managed to keep an even wall thickness all around and the core seems nicely centered. Hopefully that will mean a totally sealed volume that can be hooked up to air for some practical experiments. I have a little bank of pneumatic solenoid valves on the way, which should make for some interesting results once I’ve figured out the right pressures and plumbed the whole thing.

A few weeks back I tested out casting silicone into a corn starch based mold. Now, the molds I was using had been beaten up a bit in transit getting from Cambridge to Brooklyn. The trip rounded their corners a bit and possibly made them soak up more silicone than otherwise. Knocking around in a box on the Fung Wah bus also cracked the molds in a few spots, which is responsible for the strange shiny silicone fingers jutting off the castings in places. I haven’t done enough versions to see whether this needs to get taken into consideration when designing the cores. Chances are that, once I’ve figured out my ideal shape, I’ll have to do some thickening operation to compensate for the bleed.

So, what is the next step in approaching the robotpocalypse? Well, we’ll have to see how the tentacles hold up to some automation and control. If they act predictably and in the ways I’ve been guessing at, the next step will be picking a testing platform and generating geometry to specifically address problems as a test (like picking up and moving a ball). Otherwise, I’ll likely put my effort into designing different internal geometries to figure out how to get the distortion by air pressure to produce actions that I can predict.

The overall vision for this project is to produce a method that can generate a large variety of more or less arbitrary internal geometry, so that once a problem is defined that would need a soft robot, the robot can be designed around the constraints of the problem in the same way that brewers start out defining what kind of beer folks would like to drink, and adjusting their process until they produce a result that satisfies. The current world of experimental robotics tends to go (to keep with the beer metaphor): “I’ve made a brutally bitter IPA that tastes of licorice, mint, and coffee. Now I’m off to find people who want to buy it in kegs.” While the world rewards this kind of experimentation and exploration from time to time, I feel it’s better to fit your design to the problems the world presents, rather than coming up with a neat thing, and seeing if the world has a use for it. I accept that it’s a bit hypocritical to develop this method without a specific notion of what soft robots are good for, but I feel that I’m playing with a potentially useful technology that’s held up by who is able to play with it and how the parts get designed. Maybe some good hacking, good documentation, and an eye to making things using parts and materials that anyone can simply purchase might chip away at that barrier.