Tsinghua Team Shows Drosophila Connectome Enables Adaptable Embodied Control
Overview
A research team from Tsinghua University demonstrates that a fixed neural network core derived from the adult Drosophila connectome enables adaptable embodied control in simulated agents. The study shows that while performance is comparable to multi-layer perceptrons (MLPs) on flat terrain, the connectome-based controller significantly outperforms MLPs and other baselines when facing changes in environment (obstacles), task requirements (turning, speed variation), or body morphology (two-legged to six-legged robots). The research reveals that mechanical sensation and ascending neuron pathways are spontaneously recruited to handle these variations, suggesting that evolutionarily optimized neural structures provide reusable priors for robust motion control.
Generated from attributed reports · Updated 1 hours ago
Event evidence and corrections
1 attributed source owners. Ownership does not establish independent confirmation. Quantities are reported separately and are never added together.
Environment: simulation
“团队将成年果蝇连接组构建为固定的神经网络核心，在仿真环境中让数字果蝇进行多个维度的控制任务。”
Exact source · revision 1qbitai
Reported quantity · Reported success rate: 44.1 percent · reported trials · Differing source assertions
“以Medium迷宫为例，成功率从MLP的44.1%提高到64.2%，增加 20.1个百分点。”
Exact source · revision 1qbitai
Reported quantity · Reported success rate: 64.2 percent · reported trials · Differing source assertions
“以Medium迷宫为例，成功率从MLP的44.1%提高到64.2%，增加 20.1个百分点。”
Exact source · revision 1qbitai
Reported quantity · Reported success rate: 50.6 percent · reported trials · Differing source assertions
“对于需要6–7次转弯的复杂路线，成功率从20.4%提高到50.6%，优势扩大至约 30个百分点 。”
Exact source · revision 1qbitai
Report timeline
Follow attributed reports and material updates.
- QbitAI — Robotics on WeChatSignalTsinghua Team Shows Drosophila Connectome Enables Adaptable Embodied Control
A research team from Tsinghua University demonstrates that a fixed neural network core derived from the adult Drosophila connectome enables adaptable embodied control in simulated agents. The study shows that while performance is comparable to multi-layer perceptrons (MLPs) on flat terrain, the connectome-based controller significantly outperforms MLPs and other baselines when facing changes in environment (obstacles), task requirements (turning, speed variation), or body morphology (two-legged to six-legged robots). The research reveals that mechanical sensation and ascending neuron pathways are spontaneously recruited to handle these variations, suggesting that evolutionarily optimized neural structures provide reusable priors for robust motion control.
Event coverage history
There is not enough continuous observation data to show a trend.
Timezone · UTC
Article dates follow your selected timezone. Briefing editions use Hong Kong time (UTC+8).