←
Harsh Gupta73 posts
Harsh Gupta avatar

Harsh Gupta

@HarshGupta

Robotics @Stanford, advised by @SongShuran. Working on dexterous manipulation 🤖

◷ Joined December 2024▧ 27 Photos and videos
Jingyi Xiang avatar
⟳ Harsh Gupta retweeted
Jingyi Xiang@JingyiXiang18·Nov 11

Everyday manipulation depends on contact: plugging in a cable, unlocking a door, snapping parts together. But these skills are hard to benchmark. Tasks are diverse, and physical setups are hard to reproduce across labs. Introducing ContactBench 1.0 🧵 (1/9)

1:08
Zhanyi Sun avatar
⟳ Harsh Gupta retweeted
Zhanyi Sun@s_zhanyi·Nov 3

Can more pretraining improve a robot policy’s success rate, yet makes it worse for porttraining? 🤔 In our new work, we find that it can. Continued supervised pretraining of generative robot policies can improve reliability at the expense of coverage, and this tradeoff affects downstream posttraining. Read on to see why this happens and how we can preserve coverage. 1/n 🧵

Post attachment
Eric Ming Chen avatar
⟳ Harsh Gupta retweeted
Eric Ming Chen@ericmchen1·Oct 20

If video models scale like LLMs, they'll soon generate a perfect video plan for any embodied task. Do we need anything else? Introducing Mirror, which learns to follow these plans just by practicing, and even crafts an iron pickaxe in Minecraft w/o rewards or action labels! (1/n)

1:15
Yitang Li avatar
⟳ Harsh Gupta retweeted
Yitang Li@li_yitang·Oct 16

Can humanoids adapt to new terrain using only flat-ground references? 🤖 Meet P-BFM! 🥳 We learn a terrain-agnostic intent space with unsupervised RL and forward-backward representations, paired with a terrain-aware policy that adapts across diverse terrains—without terrain-specific demonstrations. 📄🎬💻 Check out our paper, video, and code here! 🌐 roboparty.github.io/P-BFM/

1:58
Changhao Wang avatar
⟳ Harsh Gupta retweeted
Changhao Wang@ChanghaoWang78·Oct 13

Most robot policies treat time implicitly, using sequence position as a proxy for physical time. But real robot systems rarely run on a fixed clock. Sensors run at different rates, observations arrive with delays, and actions take time to compute and execute. Successful control depends not only on what the robot observes but also on the timing of observations and actions. We introduce 🕒 Chronos Policy, a diffusion policy that explicitly conditions on observation capture times and requested action execution times. We use time-aware positional encodings to make attention depend on elapsed physical time. Because timestamps let the policy interpret each sequence's timing, we can randomize sensing and action timing during training without making those variations ambiguous. This exposes the policy to a wider range of timing conditions, improving robustness to timing changes at deployment.

0:54
Yitang Li avatar
⟳ Harsh Gupta retweeted
Yitang Li@li_yitang·Oct 9

Excited to share our work PredActor, a predictive action diffusion policy for steerable humanoid control. This direct action-generation view is appealing because it avoids the generator-tracker mismatch. 🤩

Can humanoid motion generation and control be unified in a single policy? PredActor uses a single end-to-end diffusion policy rather than separate motion generation and tracking. With proprioceptive feedback, it predicts internal future states and directly outputs actions at 50 Hz onboard a Unitree G1. Text-conditioned motion, joystick steering, disturbance response, and semantic interpolation are supported in the single policy. Project Page: masteryip.github.io/predacto…

1:17
Harsh Gupta avatar
Harsh Gupta@HarshGupta·Oct 7

🌐 Website: eigendexplore.github.io 📄 Paper: arxiv.org/abs/2610.07681 💻 Code: github.com/hgupt3/eigendexpl… Co-led with @tylerlum23, with amazing collaborators @ChanghaoWang78 and @chuer_pan, and advisors Karen Liu, @leto__jean and @SongShuran at @StanfordAILab

Harsh Gupta avatar
Harsh Gupta@HarshGupta·Oct 7

Try the interactive project page: eigendexplore.github.io/ Visualize robot rollouts in 3D, compare policies, and explore principal components learned from human hand motion

0:40
Harsh Gupta avatar
Harsh Gupta@HarshGupta·Oct 7

Beyond RL, the same idea helps sampling-based trajectory optimization. With SPIDER, we retarget human hand-and-object motion to robot hands. EigenDEXplore lowers average retargeting cost by 16% across six hands, with the same sampling budget.

0:32
Harsh Gupta avatar
Harsh Gupta@HarshGupta·Oct 7

EigenDEXplore also helps functional retargeting: reproducing human-demonstrated object motion with two robot hands With DexMachina's reference-guided RL, it improves average object-tracking accuracy by 28% with task reward alone

0:20
Harsh Gupta avatar
Harsh Gupta@HarshGupta·Oct 7

These gains transfer to the real robot On DeXtreme cube reorientation, mean consecutive successes rise from 8.45 to 11.90 (+41%) over 40 trials per method Human-guided exploration during training, full joint-space control at deployment

0:31
Harsh Gupta avatar
Harsh Gupta@HarshGupta·Oct 7

Cube reorientation can require motions outside a human-derived subspace Eigen-Space restricts actions to that subspace and struggles EigenDEXplore uses it to guide exploration, keeping full joint-space actions: 40% more consecutive successes than Joint-Space in sim

0:36
Harsh Gupta avatar
Harsh Gupta@HarshGupta·Oct 7

What changes when exploration is coordinated? EigenDEXplore discovers useful grasps and learns faster in this tool-manipulation comparison. On real-world tool use, average task progress rises from 33% to 82% across hammering, wiping, scooping, and brushing.

0:33
Harsh Gupta avatar
Harsh Gupta@HarshGupta·Oct 7

Dexterous manipulation requires coordinated joint motion, but standard RL often explores with independent joint noise EigenDEXplore combines independent joint noise with noise along "eigen" directions: coordination patterns extracted from human motion using PCA

0:40
Harsh Gupta avatar
Harsh Gupta@HarshGupta·Oct 7

Per-joint Gaussian noise is a terrible way to explore a 20+ DoF hand We introduce EigenDEXplore, which explores along coordinated patterns from human hand motion, while keeping full control of every joint Better learning, from tool use 🔨 to reorientation 🔄 to bimanual 👐

0:54