AI Enables Full-Body Object Manipulation by Robots

MIT uses reinforcement learning to enhance robots' full-body manipulations, overcoming computational challenges during contact.

[{"selector":"#anim-11dfb674-a618-4432-a9f7-a9d432095dfb","keyframes":{"transform":["translate3d(-116.07717%, 0px, 0)","translate3d(0px, 0px, 0)"]},"delay":0,"duration":600,"easing":"cubic-bezier(0.4, 0.4, 0.0, 1)","fill":"both"}] [{"selector":"#anim-1ef3062b-4316-46e6-a06d-657d4ff971d9","keyframes":{"opacity":[0,1]},"delay":0,"duration":600,"easing":"cubic-bezier(0.4, 0.4, 0.0, 1)","fill":"both"}] [{"selector":"#anim-e30701ae-216c-4fe2-9f0a-1f595a4508f2","keyframes":{"transform":["scale(0.15)","scale(1)"]},"delay":0,"duration":600,"easing":"cubic-bezier(0.4, 0.4, 0.0, 1)","fill":"forwards"}]

Smoothing specific model equations enables robots to benefit from reinforcement learning without extensive trajectory calculations.

[{"selector":"#anim-965cb874-e428-455e-b147-2aabda1ad3e4","keyframes":{"transform":["translate3d(-116.39871%, 0px, 0)","translate3d(0px, 0px, 0)"]},"delay":0,"duration":600,"easing":"cubic-bezier(0.4, 0.4, 0.0, 1)","fill":"both"}] [{"selector":"#anim-d747f4b8-32f5-4160-9499-3aac49e4993a","keyframes":{"opacity":[0,1]},"delay":0,"duration":600,"easing":"cubic-bezier(0.4, 0.4, 0.0, 1)","fill":"both"}] [{"selector":"#anim-733e56af-d21f-46e9-beec-f6bc41633b19","keyframes":{"transform":["scale(0.15)","scale(1)"]},"delay":0,"duration":600,"easing":"cubic-bezier(0.4, 0.4, 0.0, 1)","fill":"forwards"}]

Potential applications include smaller, mobile robots for manufacturing and adaptable space exploration using onboard computers.

[{"selector":"#anim-a1cb1320-61d3-400c-b700-89ef9786b262","keyframes":{"transform":["translate3d(-116.29393%, 0px, 0)","translate3d(0px, 0px, 0)"]},"delay":0,"duration":600,"easing":"cubic-bezier(0.4, 0.4, 0.0, 1)","fill":"both"}] [{"selector":"#anim-4e7600d2-565f-400e-b45f-7cb57c79ee40","keyframes":{"opacity":[0,1]},"delay":0,"duration":600,"easing":"cubic-bezier(0.4, 0.4, 0.0, 1)","fill":"both"}] [{"selector":"#anim-93d6ef76-c360-465b-a6b1-043cb4a2fef2","keyframes":{"transform":["scale(0.15)","scale(1)"]},"delay":0,"duration":600,"easing":"cubic-bezier(0.4, 0.4, 0.0, 1)","fill":"forwards"}]

Efficient simulations show comparable performance and reduced computation time, validated in real-world tests on robotic arms.

[{"selector":"#anim-e05c9116-5e2e-4104-a608-d7ca9e551ffc","keyframes":{"transform":["translate3d(-115.2381%, 0px, 0)","translate3d(0px, 0px, 0)"]},"delay":0,"duration":600,"easing":"cubic-bezier(0.4, 0.4, 0.0, 1)","fill":"both"}] [{"selector":"#anim-da0addab-77f8-4eaf-8511-31ae5fc4a39c","keyframes":{"opacity":[0,1]},"delay":0,"duration":600,"easing":"cubic-bezier(0.4, 0.4, 0.0, 1)","fill":"both"}] [{"selector":"#anim-3b701fb7-2cd0-42a7-9b24-2e48472d3672","keyframes":{"transform":["scale(0.15)","scale(1)"]},"delay":0,"duration":600,"easing":"cubic-bezier(0.4, 0.4, 0.0, 1)","fill":"forwards"}]

Future improvements aim to extend the technique for dynamic motions, closing the performance gap between robots and humans in physical tasks.

[{"selector":"#anim-21834f99-401f-4c52-8082-e1a406e5985b","keyframes":{"transform":["translate3d(-115.55555%, 0px, 0)","translate3d(0px, 0px, 0)"]},"delay":0,"duration":600,"easing":"cubic-bezier(0.4, 0.4, 0.0, 1)","fill":"both"}] [{"selector":"#anim-776e7dc5-9e66-43e1-b533-7637d84d24c3","keyframes":{"opacity":[0,1]},"delay":0,"duration":600,"easing":"cubic-bezier(0.4, 0.4, 0.0, 1)","fill":"both"}] [{"selector":"#anim-d4c79871-355d-443c-800c-013d5d63751b","keyframes":{"transform":["scale(0.15)","scale(1)"]},"delay":0,"duration":600,"easing":"cubic-bezier(0.4, 0.4, 0.0, 1)","fill":"forwards"}]